INTERNET-DRAFT K. Miller, StarBurst
Expires in Six Months K. Robertson, StarBurst
A. Tweedly, Cisco
M. White, StarBurst
January, 1997
StarBurst Multicast File Transfer Protocol (MFTP) Specification
<draft-miller-mftp-spec-01.txt>
Status of this Memo
This document is an Internet-Draft. Internet-Drafts are working
documents of the Internet Engineering Task Force (IETF), its
areas, and its working groups. Note that other groups may also
distribute working documents as Internet-Drafts.
Internet-Drafts are draft documents valid for a maximum of six
months and may be updated, replaced, or obsoleted by other
documents at any time. It is inappropriate to use Internet-
Drafts as reference material or to cite them other than as
ôwork in progressö.
To learn the current status of any Internet-Draft, please check
the ``1id-abstracts.txt'' listing contained in the Internet-
Drafts Shadow Directories on ftp.is.co.za (Africa),
nic.nordu.net (Europe), munnari.oz.au (Pacific Rim),
ds.internic.net (US East Coast), or ftp.isi.edu (US West Coast).
Abstract
The Multicast File Transfer Protocol (MFTP) is a protocol that
operates above UDP in the application layer to provide a reliable
means for transferring files from a sender to up to thousands
(potentially millions with network "aggregators" or relays) of
multiple receivers simultaneously over a multicast group in a
multicast IP enabled network. The protocol consists of two parts; an
administrative protocol to set up and tear down groups and sessions,
and a data transfer protocol used to send the actual file reliably and
simultaneously to the multiple recipients residing in the group .
Intellectual Property Rights
StarBurst Communications owns U.S Patent 5,553,083 covering the method
for data transfer used in the Multicast File Transfer Protocol (MFTP)
described herein. StarBurst hereby records a grant by StarBurst
Communications Corporation to permit the conditional free use of this
patent to help facilitate adoption of MFTP by the Internet community
StarBurst hereby covenants that it will not assert its U.S. Patent
5,553,083 or its non-U.S. counterparts against any party that
implements MFTP or any derivative that includes MFTP, and operates in
compliance with MFTP or such derivative, when MFTP or such derivative
is employed in connection with any Internet Protocol.
Miller, et. al. [Page 1]
INTERNET-DRAFT MFTP Specification January 1997
This grant of rights will terminate with respect to any party that
asserts a patent it owns, directly or indirectly, against StarBurst
for StarBurstÆs implementation of and operation in compliance with
MFTP or any derivative, as defined above. The termination will occur
as of the date the patent is asserted against StarBurst.
A written confirmation of this grant, and/or a license under any other
StarBurst patent under StarBurstÆs then current terms, conditions, and
royalty rates must be obtained by sending a request to:
Edward M. Durkin
StarBurst Communications Corp.
150 Baker Avenue
Concord, MA 01742
Table of Contents
1. Introduction.....................................................3
2. Definitions......................................................5
3. MFTP Architecture................................................8
4. Scaling..........................................................10
4.1 Scaling Extensions.........................................11
5. Performance......................................................11
6. Group Management.................................................11
6.1 Joining Clients to Private ............................26
11.2.1 File Delivery - Server..............................26
11.2.2 File Delivery - Client..............................29
Miller, et. al. [Page 2]
INTERNET-DRAFT MFTP Specification January, 1997
11.3 Product Completion........................................30
11.3.1 File Product Completion - Client....................31
11.3.2 File Product Completion - Server....................32
11.4 Flow Control..............................................34
12. Message Formats.................................................34
12.1 MCP Messages .............................................37
12.1.1 Query Group Message..................................38
12.1.2 Join Group Message ..................................39
12.1.3 Leave Group Message .................................40
12.1.4 Echo Message ........................................40
12.1.5 Echo Response Message ...............................41
12.2 MDP Messages..............................................41
12.2.1 Announce Message.....................................41
12.2.2 Registration Message ................................47
12.2.3 Data Transfer Message ...............................48
12.2.4 Status Request Message...............................48
12.2.5 NAK Response Message.................................50
12.2.6 ACK Response Message.................................51
12.2.7 Done Message.........................................52
12.2.8 Completion Message...................................53
12.2.9 Abort Message........................................54
12.2.10 Quit Message........................................54
12.2.11 End Message.........................................54
13. Reason 1. Simplicity, which generally means increased reliability
2. Virtually no performance change over high delay networks such as
satellite
3. Ability to be scaleable to thousands of recipients over one hop
networks such as satellite with no intermediate relaying
entities
4. Ability for the sender (Server) to have complete knowledge of
the state of receivers (Clients)
Additionally, MFTP has flexibility so that optional features can be
invoked to increase scalability beyond thousands, including:
1. Aggregation and relaying of responses by the network
infrastructure or other intermediate entity
2. Response suppression to reduce responses from members of a
subnet similar to that used by IGMP as specified in RFC 1112
Both of these techniques can be used to increase scalability by
reducing Client message traffic to the Server.
Typical applications for MFTP include electronic software
distribution, transmission of critical information to field offices,
distribution of multimedia information to local servers, and
replication of web servers to the edges of networks for improved
performance. A significant new application is the ability to provide
subscription based "push" information delivery to receivers who have
signed up for a particular information service.
MFTP is designed to operate with networks that support multicast and
broadcast data transport at the link layer such as LANs and SMDS, and
with multicast IP router networks. Multicast user groups can be set
up for either type network. When the network is router-based, data is
delivered only to hosts in the multicast group based on multicast IP
addresses. Each host supports RFC1112, Host Extensions for IP
Multicasting.
When the network does not support multicast, data may be broadcast
with multicast groups defined at the application layer at each MFTP
host. This means that data is broadcast within the network but
filtered at the MFTP host so that only members of the defined group
receive the data. Data may also be sent in unicast mode to a single
receiver.
The MFTP Server (data sender) manages the multicast groups, initiates
the file transfer, and controls the transfer operation. The MFTP
Miller, et. al [Page 4]
INTERNET-DRAFT MFTP Specification January, 1997
Client (data receiver) joins the multicast group, receives the data
sent by the Server, and provides reception status to the Server when
requested.
MFTP transmission is efficient because data is sent in a streaming
mode. This means that the Server continuously sends data without
waiting for responses from Clients. Acknowledgment traffic generated
by the Clients is minimized because the Clients send responses only
for Data Transmission Units (DTUs) that are not received. In addition,
the status for a range of DTUs is aggregated in each response, further
reducing the response traffic. At the end of a file transmission, the
DTUs reported by the Clients as lost or received in error are
retransmitted by the Server.
MFTP supports file checkpoint/restart so that an interrupted file
transfer may be resumed without retransmitting data that has already
been received by the Clients. Unlike other protocols that use all
available bandwidth to transmit data, MFTP allows a maximum Server
transmit rate to be specified. This allows files to be transmitted
without stealing bandwidth needed by other applications.
MFTP consists of two protocol components - the Multicast Control
Protocol (MCP) and the Multicast Data Protocol (MDP). MCP allows the
Server to dynamically control the joining and leaving of Clients to
multicast groups. MDP then handles the reliable transmission of data
products to the Clients that have joined the multicast group.
The major features and characteristics of MFTP are summarized below.
Reliable data transfer via selective NAK mechanism
Uses multicast transmission, as defined in RFC 1112
Scaleable to thousands of recipients without intermediate entities
Scaleable to millions with intermediate entities to aggregate
responses
Includes multicast group management and control
First level congestion control mechanism defined
Minimal performance change on high delay networks such as satellite
Additionally supports broadcast and unicast transmission
The maximum data rate of each transmission is specified
File checkpoint/restart is supported
2. Definitions
This section provides definitions for terms used in this document.
Aggregation: The process of collecting responses from Clients and
combining them into one response to be relayed back to the Server.
This reduces the back traffic to the Server and thus increases the
number of Clients that can be served.
Announcement: The process of informing Clients that a data transfer is
about to start. This involves sending a message to the "well known"
multicast group address where Client hosts have joined and are
listening for such messages. The message identifies the data product,
Miller, et. al. [Page 5]
INTERNET-DRAFT MFTP Specification January, 1997
its size, name, etc. This message is retransmitted at regular
intervals for a configured time period (e.g. 3 minutes). After the
product announcement phase is completed, the Server begins the
transmission of the product (see Product Delivery below).
ARS: Application Reference String. This is a variable length, case-
sensitive string that the application provides to identify itself. It
is used by MFTP to ensure that data is sent and received by like
applications.
Authorized Client: A term used primarily for Closed Groups where the
IP address of each Client that is allowed to receive the data is
included in the Announce message. That is, the Client is "authorized"
by the applications to receive the data if the Client finds its IP
address in the Announce message. All other Clients are prohibited from
receiving the data.
Block: A group of MFTP data messages. Clients report the receive
status of data messages a block at a time.
Client: The receive function of the MFTP protocol. A single MFTP
implementation may contain only a Client function, or it may contain
both a Client and Server function (see Server below). Therefore,
Client does not necessarily refer to the host itself.
Closed Group: A form of group management where the Server specifies
the Clients that are allowed to participate in the data transfer. All
other Clients are prohibited from receiving the data. Also refer to
Open Limited Group and Open Unlimited Group.
Data Product: A file that is sent by the Server to Clients.
Done Message: A message sent from the Clients to the Server to
indicate that the Client has received a complete data product.
DTU: Data Transfer Unit. A protocol message which contains the data
that is being sent by the Server to the Clients.
Message Implosion: With a large number of Client hosts, all sending
messages to the Server, it is possible for the Server to be overrun
and messages to be lost. It is important for the protocol to minimize
the number of messages that are being directed to any single host. An
example of this is the negative-acknowledgment scheme that is used to
obtain a Client's reception status and the way that the status for a
range of DTUs is grouped together into a single response message.
MFTP: Multicast File Transfer Protocol. The protocol described in this
document.
MTU: Maximum Transmission Unit. The largest unit of data, in bytes,
that can be transmitted over the user's physical network.
Open Limited Group A form of group management where the Server does
Miller, et. al. [Page 6]
INTERNET-DRAFT MFTP Specification January, 1997
not specify which Clients may participate in the data transfer. Any
Client that wishes to receive the data is required to register with
the Server. The Server may limit the number of participants based on
its own resources or some other criteria. Also refer to Closed Group
and Open Unlimited Group.
Open Unlimited Group: A form of group management where the Server does
not specify which Clients may participate in the data transfer.
Clients that wish to receive the data are not required to register
with the Server. Also refer to Closed Group and Open Limited Group.
Pass: For a file transmission, MFTP makes one "pass" through the file
to transmit the file data. It then makes another "pass" through the
file to retransmit data that was not received during the first pass.
Additional passes may also be made to deliver the product successfully
to all receiving hosts.
Private Address: The network address to which a Server transmits the
data product. The Server specifies the Private Address in the Announce
message for each product transfer. The private address is most
relevant in multicast transmission because separate multicast
addresses are used for product announcement and delivery. Only Clients
that are authorized to receive a product actually join the multicast
group defined by the private address.
Product Delivery: The process of transmitting the data product from a
Server to Clients. The data is transmitted in messages called Data
Transfer Units (DTUs). A group of DTUs is called a block. The
transmission of the entire data product once is called a pass. No DTU
retransmissions occur during a pass. Rather, additional passes are
made when retransmissions are required. However, only the missed DTUs
are retransmitted on each subsequent pass. This phase is always
preceded by the Product Announcement phase (see Announcement above).
Public Address: The network address to which a Server transmits a
product announcement. The public address is most relevant in multicast
transmission because separate multicast addresses are used for product
announcement and delivery. Any number of Servers may announce products
to a single public address and any number of Clients may join the
multicast group defined by the public address.
Registration: The process of a Client informing a Server that the
Client intends to participate in the product transmission that is
currently being announced. The Client does this by sending a
Registration response message to the Server. This response is sent to
the address specified by the "Response Address" parameter in the
Announce message.
Relay: An entity that forwards messages to another destination after
receiving them. For example, an aggregator also acts as a relay by
combining Client responses and then forwarding them to the Server or
another aggregator. Relays can also be Clients that in turn act as
Servers for another group that cannot be reached directly by the
Server or for efficiency.
Miller, et. al. [Page 7]
INTERNET-DRAFT MFTP Specification January, 1997
Response Address: An IP address (unicast or multicast) to which
Clients send Registration, Status, and Done messages. The Response
Address is specified in the Announce message. The ability to specify
an address other than the Server's address allows an intermediate
entity to perform aggregation/relaying of the responses.
Server: The transmit function of the MFTP protocol. A single MFTP
implementation may contain only a Server function, or it may contain
both a Server and Client function (see Client above). Therefore,
Server does not necessarily refer to the host itself.
Transmission ID: A number that uniquely identifies an individual data
product transmission that originates from a single Server. The
concatenation of the Server's IP address and the Transmission ID
uniquely identifies any MFTP transmission in the network. The
Transmission ID is included in every MDP message sent by the Server
and Client.
3. MFTP Architecture
MFTP operates above the User Datagram Protocol (UDP) layer of the
TCP/IP protocol suite. As such, it uses the UDP Sockets interface.
MFTP defines both a send function (the Server) and a receive function
(the Client). The Server only transmits data products and the Client
only receives data products. An implementation may optionally include
both functions, as shown in Figure 3-1 or may only have one. The
checksum, defined in UDP, is the usual mechanism to determine if a
datagram is corrupted. Some implementations may not include a UDP
checksum; in these cases, an optional checksum is provided by MFTP.
+-+-+-+-+ +-+-+-+-+
| File | | File |
+-+-+-+-+ +-+-+-+-+
| +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ |
+-+-+-+->| Server | Client |+-+-+-+->
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ^
V |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| UDP Layer |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| IP Layer |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 3.1 MFTP Data Architecture
File products can be transmitted directly from the file system by the
Server and written directly to the file system by the Client. All
implementations are required to support the MFTP Echo message and Echo
Response message, which provide functions similar to "ping".
Miller, et. al. [Page 8]
INTERNET-DRAFT MFTP Specification January, 1997
Transmission Modes
Messages sent by the Server may be sent in either unicast, broadcast,
or multicast mode. The Client must be able to receive Server messages
in any of these modes. This allows flexibility in designing the Server
to achieve network optimization goals. For example, unicast may be
used to send an Abort message to a single Client when that is the only
Client that is being aborted. All other Clients are spared from having
to receive the message on the multicast group. An entire product
transfer can occur in unicast mode when there is only a single Client
receiver.
Broadcast mode may be used when the network does not support IP
multicast. All Clients receive the messages, but messages that are not
directed to the Client are filtered and discarded. This mode should
normally be avoided because of its impact on the network.
The remainder of this document generally assumes multicast
transmission, unless otherwise stated.
Well-Known UDP Port
IANA-assignment of a "well-known" UDP port will be requested for MFTP.
Certain MFTP messages must be sent to this port because it will be the
only port number known both to the sender (Server) and the receivers
(Clients). This includes the MCP messages, the Announce message and
the Data message.
An MFTP station may include both the Server and Client function, both
using the well-known port. To reduce the traffic on the well-known
port, a Server may dynamically allocate a source "session" port to
send a data product and to receive responses from the Clients. The
Announce message is sent to the Client's well-known port. Client
responses are returned to the source session port number. The Client
may also allocate a session port to send messages, rather than using
the well-known port.
The Server sends all messages to the Client's well-known port. The
specified session port is in effect only for the current product
delivery. This use of session ports is illustrated in Figure 3-2.
Miller, et. al. [Page 9]
INTERNET-DRAFT MFTP Specification January, 1997
+-+-+-+-+-+ +-+-+-+-+-+
| | | |
| +-+-+-+-+-+-+-+-+ Announce, data+-+-+-+-+-+-+-+-+ |
| |well known port| /+-+-+-+-+-+->|well known port| |
| +-+-+-+-+-+-+-+-+ / +-+-+-+-+-+-+-+-+ |
| Server | / | Client |
| | / | |
| +-+-+-+-+-+-+-+-+/ Client Resp. +-+-+-+-+-+-+-+-+ |
| | session port |<-+-+-+-+-+-+-+-+-| session port | |
| +-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+ |
| | | |
+-+-+-+-+-+ +-+-+-+-+-+
Figure 3-2 Session Ports
4. Scaling
By focusing on file based delivery rather than attempting to be a
generalized reliable multicast transport layer, improvements in
scaling without requiring intermediate aggregation points are
achieved. It is known that the limitation for scaling in reliable
multicast protocols is the ôNAK Implosionö problem, where the
administrative (NAK) back traffic from receiver (Client) to sender
(Server) eventually overwhelms the sender as the number of receivers
grows. The use of ôblocksö in MFTP aggregates NAKs from each
recipient so that one NAK with a bit map of the DTUs in the block can
represent thousands of bad or dropped DTUs, depending on the actual
number of bad or dropped DTUs in the block.
Although not required, MFTP recommends that the block size be tied to
the MTU size so that the bit map in one NAK packet exactly fits the
number of DTUs in a block. For example, if the MTU is 1500 bytes at
the MAC layer (the maximum for Ethernet), the block size consists of
over 11 thousand DTUs. Thus, a burst of thousands of dropped DTUs
within a block results in only one NAK rather than thousands (see
Section 11.2.1).
Even when the MTU at the IP layer is 576 bytes, the maximum guaranteed
in TCP/IP networks with no fragmentation, the block size is over 3800
DTUs which still provides a high level of minimization of NAK traffic.
Experiments have been performed by StarBurst to emulate up to 10,000
Client receivers in one Closed group transmission and no significant
performance degradation was observed. Obviously, scaling behavior
will be affected by percent error rate and/or packet loss rate in the
network, with the better the network characteristic the better the
scaling. With MFTP, the Open Unlimited group model has scaling
performance totally dependent on NAK traffic; however, for Closed and
Open Limited group models, all Client receivers Register and all
Client receivers send Done messages, the latter acting as one ACK for
the entire file. Thus, in these models, all receivers send one
message at the beginning and at the ending of the transmission. This
Miller, et. al. [Page 10]
INTERNET-DRAFT MFTP Specification January, 1997
traffic in fact becomes the scaling limitation in the Closed and Open
Limited group models.
Products using MFTP in essentially the same form as described in this
document have been operating in private networks since mid 1995. At
the time of this writing, several installations are operating in a
production environment with over 1000 simultaneous receivers and one
installation expects to be operating with about 9000 receivers in a
group later in 1997.
4.1 Scaling Extensions
In addition to the minimization of administrative back traffic
described above, provision is made in MFTP to extend scaling to
hundreds of thousands and perhaps millions of simultaneous receivers
depending on the network characteristics by using network aggregation
entities to further consolidate administrative back traffic.
Additionally, provision is made for NAK suppression on subnets,
similar to the method used by IGMP as specified in RFC 1112.
5. Performance
MFTP is a rate based protocol with a transmission rate defined by the
sender (Server). By design, it does not attempt to provide error
correction in real time but takes advantage of the non-real time
nature of file transmission. Thus, transmission remains continuous at
the set transmission rate through the total transmission of the file
in the first ôpassö, with corrections made in subsequent passes. This
makes MFTP very insensitive to delay such as occurs over satellite
networks and also makes it tolerant of asymmetric data channels.
Performance is affected as the number of simultaneous receivers grows,
as this increases the number of bad or dropped packets in aggregate
for the group resulting in more retransmissions. However, in practice
there is often high correlation in bad/dropped packets among
receivers. For example, all receivers downstream from a congested
node will experience the same packet drops caused by that node. Since
the retransmissions are sent to the same multicast group, they will
all get the same correction simultaneously on a subsequent pass.
Protocol overhead is modest, both in header size and in the amount of
processing needed. Thus, MFTP can be expected to be able to exceed
the performance of FTP even in a point to point application.
6. Group Management
MFTP uses the push method for distributing data products. That is, at
a time determined by the Server user application, a data product
transmission is "announced" to the Client population. This
announcement consists of transmitting Announce messages for a
period of time, followed by the data product itself. The Announce
message identifies the data product that will be transmitted, its
size, and other parameters documented herein.
Miller, et. al. [Page 11]
INTERNET-DRAFT MFTP Specification January, 1997
MFTP uses two separate multicast groups, called the public group and
the private group. The Server announces products on its public group
address and Clients join the address to receive announcements. The
product data is transmitted on the private group address. That is, the
public group address is where any Client may join and listen for
announcements made by any Server that is sending to the public
address. However, only Clients that are authorized by the application
to receive a particular transmission actually join the private
address.
The public/private group architecture is shown in Figure 6-1. Clients
1 - 3 are joined to and listening on the public group. Only Client 3
wants to receive the product that has been announced and has joined
the private group to receive it.
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+ +-+-+-+-+
| Server |+-+-+-+-+-+->| Public Group |+-+-+-+-+-+->|Client1|
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+\ +-+-+-+-+
| \ \
| \ \
| \ \ +-+-+-+-+
| \ \+-+-+-+>|Client2|
| \ +-+-+-+-+
| \
| +-+-+-+-+-+-+-+-+ \+-+-+->+-+-+-+-+
+-+-+-+-+-+-+-+-+-+-+>| Private Group |+-+-+-+-+-+->|Client3|
+-+-+-+-+-+-+-+-+ +-+-+-+-+
Figure 6-1 Group Architecture
Group management deals with how Clients learn about and are joined
to the public and private groups. The attributes of the public group
are discussed first, with the assumption that Clients are already
joined to it. Once a Client is joined to the public group, it learns
about the private group in the Announce message. Finally, the method
used to join Clients to the public group is discussed.
6.1 Joining Clients to the Private Group
Use of the public group is completely flexible in that there may be
one for each Server or multiple Servers may transmit product
announcements to a single group address. It is recommended that the
number of public groups be kept to a minimum, even one. This allows
the use of a single public group for Clients to listen for
announcements from many Servers without requiring a separate multicast
group for each one.
In order to direct the product data to only those Clients that
actually want to receive it, the product announcement message includes
a field that specifies a separate private multicast address. The
Clients then dynamically join the private address and the Server
transmits the product data on that address. Note that while joined to
Miller, et. al. [Page 12]
INTERNET-DRAFT MFTP Specification January, 1997
the private group the Client continues to be a member of and listens
for product announcements on the public address. This allows the
Client to receive multiple data products simultaneously, if it so
desires.
Normally, the Clients leave the private address at the end of the
product transfer and continue to listen for product announcements
on the public address. However, if the Server is going to transmit
multiple data products, it may set an option in the product
announcement that directs the Clients to not leave the private
address. This reduces the overhead associated with joining and leaving
multicast groups. By also announcing the next product transmission
on the public group, Clients that did not hear or did not participate
in the previous transfer may be added to the private group.
If the Server is supplied with more than one private address, it may
arbitrarily use any one of the addresses to transmit a data product.
Multiple addresses would typically be used to send multiple products
to different groups at the same time by a single Server.
MFTP includes a unique transmission identifier in each message so that
the Client can always distinguish messages that are associated with an
individual data transmission.
6.2 Joining Clients to the Public Group
To discuss how Clients are joined to the public address, there are two
models that are considered, Closed and Open.
In the Closed model, the Server knows in advance which Clients
are authorized to receive a data transmission, and the number of
Clients is relatively small (e.g. thousands of Clients). This allows
the Server to tightly control group membership.
The public address may be known and configured in advance at the
Clients or the Server may use the MFTP Multicast Control Protocol to
direct Clients to join the public group. This type of group
management is called Closed Groups since the Server restricts the
membership of the group. This control occurs at two levels - (1)
either by configuration or by use of MCP, the members of the group
are restricted and (2) the product announcement itself includes a list
of Clients that are authorized to participate.
Upon receiving the Announce message, the Clients designated to join
the group register with the Server by sending a Registration message.
All other Clients are administratively told not to join the group and
thus do not receive the transmission. This is a Server-centric mode of
operation that allows the Server to tightly control which Clients
receive the data and also minimizes the number of Registration
messages arriving at the Server. It also allows the Server to keep
detailed statistics on each Client. This mode is most useful when the
data product has a value associated with it that prevents it from
being freely distributed.
Miller, et. al. [Page 13]
INTERNET-DRAFT MFTP Specification January, 1997
In the Open models, the receiving Clients are not known to the Server.
This means that the Server is not able to configure the multicast
groups via MCP. A Client must be able to obtain the public group
address of Servers that it is interested in, perhaps by using the MCP
Group Query message or by using one of the external protocols defined
in the mmusic working group.
Since the MFTP server is not determining those Clients that join its
session, this type of group management is called "Open Group".
There are two variations of Open Groups, called "Limited" and
"Unlimited", as discussed below.
The Limited mode allows the Server to announce a transmission without
specifying the Clients that are authorized to receive the data. This
may be because the Server does not know in advance which Clients may
want to receive the data, or the number of potential Clients may be
very large.
This type of group allows any Client to request participation in the
data delivery, but the Server limits the number of actual participants
based on some criteria determined by the application. As Clients
register to receive the data, the Server learns the identity of each
Client. Therefore, like the Closed Group, the Server is able to keep
detailed statistics on each Client. In fact, after the
Announce/Registration phase, protocol operation is identical to the
Closed Group. This mode is most useful when it is permissible to
freely distribute the product, but the sender wants to know who is
receiving it.
The Unlimited mode allows any Client to participate in the data
delivery and the Server does not limit the number of participants. To
achieve this, certain types of Client responses are waived to prevent
message implosion at the Server. This includes the Registration and
Done messages. The only Client response that is retained is the Status
Response message. Since Clients send this message only for DTUs that
are not received, there will not necessarily be a response from every
Client. The Server is unable to identify each receiving Client when
this mode is used, so it is most useful when the data product can be
freely distributed and the sender does not care to know who is
receiving the product.
7. Response Address
The Server specifies the response address parameters in the Announce
message. Each Client then sends all response messages to the specified
response address (regardless of whether the Server message is received
in unicast or multicast mode).
A different response address than the Server may be used by an
aggregating entity in the network to collect responses from clients
and reduce implosion traffic to the Server. This increases the
scalability, i.e., the number of clients supported. The use of a
multicast response address with a suitable TTL may be used for
Miller, et. al. [Page 14]
INTERNET-DRAFT MFTP Specification January, 1997
aggregation and is required for response suppression (see Section 6).
The response parameters include the following:
Response IP Address (may be different than Server's IP address)
Response port number (may be different than Server's port number)
Server IP address
Server port number
The Response IP Address may be either a unicast or multicast address.
The normal mode is that the Response IP Address and port number be
that of the Server if optional aggregation and/or suppression is not
used.
The Server IP address and port number are echoed in the Client's
response. This is the address and port where the response must
eventually be sent by an intermediate relay point, if present.
8. Response Suppression
Response message suppression is an optional Client function that
eliminates the sending of redundant status messages from a single
subnet, such as a LAN. That is, congestion that occurs upstream from
the subnet will produce an identical set of dropped messages at each
Client on the subnet. The result is all Clients send identical status
responses to a Status Request message. This can be avoided by having
the Clients listen to each other's response and suppress duplicate
responses.
Response message suppression is used only with the sending of the NAK
Response message. Other Client responses are sent in the normal manner
described in this document.
The NAK Response message is sent to the multicast Response Address
as specified in the Announce message. Each Client must be joined to
this address in order to hear the response of other Clients on the
subnet.
When a Status Request message is received by the Client, and the
request causes a NAK response message to be built, the Client does not
immediately send the message. Rather, a delay timer is started. Each
Client's delay timer is a different, randomly-chosen value between 0
and n seconds, where n is configurable with a default value of 1
second. The Client uses a pseudo-random number generator to compute
its delay timer value, using the Client's own host IP address as part
of the seed to reduce the chance of multiple Clients generating the
same delay.
While waiting for its own delay timer to expire, each Client receives
and examines the status responses sent by other Clients. If the pass
number, block number, and event hash fields all match the Client's
unsent message, the Client's unsent message is discarded and the delay
timer is canceled. Otherwise, the Client continues to listen to and
examine additional responses. If no matching message is received
Miller, et. al. [Page 15]
INTERNET-DRAFT MFTP Specification January, 1997
before the Client's delay timer expires, the Client transmits its own
response message. Any response messages received after the Client's
message has been transmitted are ignored. If a new Status Request
message is received from the Server while the delay timer is running,
the Client immediately sends its current response message, cancels the
timer, and processes the new Status Request as described above.
The response suppression function is applied at the subnet level.
Since all Clients participating in a product transfer will join to the
multicast Response Address, messages that leave a subnet will also
arrive at all other Clients joined to the group and will be processed
as described above. When message aggregation is performed by the
network infrastructure (see Message Aggregation above), the network
can prevent responses from being propagated anywhere except back to
the Server.
9. Configuration
The following configuration parameters affect the operation of the
protocol:
Announce Time Limit
Fast Announce Option
Announce Interval
Announce Persistence
Status Interval
Status Retry Timer
Status Retry Count
Delivery Time Limit
Completion Interval
Transmit Rate
Register Retry Timer
Register Retry Count
Response Back-off Timer
Response TTL
Some parameters may apply only to product transmission, only to
product reception, or to both. The application of each parameter is
noted in the following descriptions. Also refer to the description of
the protocol and message formats below for additional information.
9.1 Announce Time Limit
This parameter sets the maximum time, in seconds, that MFTP will
use for the announce phase of a product transmission. The announce
phase will last for this duration unless the Fast Announce Option (see
below) is used.
Miller, et. al. [Page 16]
INTERNET-DRAFT MFTP Specification January, 1997
9.2 Fast Announce Option
This parameter only applies to Closed Groups. The Server may shorten
the duration of the Announcement immediately after all members of the
group register. This is the normal mode of operation in the Closed
Group model.
9.3 Announce Intervals
This parameter sets the interval, in seconds, between successive
transmissions of the Announce message during the product announce
phase.
9.4 Announce Persistence
This parameter specifies whether the Announce message should
continue to be sent during the product delivery phase. This could be
desirable in the Open Unlimited group model where the Announce can act
as an ôadvertisement" for what is being transmitted.
9.5 Status Interval
This parameter specifies the interval, in seconds, between Status
Request messages sent by the Server to the Clients. It is used to
establish the transmit block size. A default block size will be
computed, based on the MTU. This block size will maximize the amount
of NAK information that can be sent in a single status response
message by the Client. The Server sends the status request message at
the end of each transmitted block. However, for a large MTU (and
therefore a corresponding large block size), a relatively long time
may elapse before there is any status information that can be returned
to the application about the NAK status of individual Clients. The
application may set the Status Interval parameter to the time interval
that it wants the Server to request NAK status from the Clients. The
block size is adjusted for the requested interval so that a block of
DTUs is sent in the interval and then NAK status is requested by the
Server.
9.6 Status Retry Timer
This parameter specifies the time duration, in seconds, that the
Server will wait for Client responses after a Status Request message
is sent.
This timer is not used for Status Request messages sent during a pass
because the Server does not stop and wait for responses. However, at
the end of the pass, the Server cannot begin the next pass if no
responses (status response or Done message) have been received for
the current pass. In this case, the Server sends a Status Request
message, starts the Status Retry timer, and waits for any responses.
When the timer expires, the Server begins the next pass if any status
response messages were received. If there is no response, the Server
sends another Status Request message and restarts the retry timer.
Miller, et. al. [Page 17]
INTERNET-DRAFT MFTP Specification January, 1997
When there are no responses, this procedure continues up to the
retry limit set by the Status Retry Count (see below) or until the
Delivery Time Limit is reached, whichever occurs first.
9.7 Status Retry Count
This parameter specifies the maximum number of times that the
Server will retransmit a Status Request message when no response is
received from any Client. Refer to Status Retry Timer above to see
how it is used.
9.8 Delivery Time Limit
This parameter sets the maximum time for the product delivery phase of
a file transmission. The parameter is an integer that represents a
percentage, with a minimum value of 100% (i.e. value =100). MFTP
multiplies the integer by the optimal time that it takes to make one
pass through the transmit file. The result is the maximum time devoted
to the product delivery phase, in seconds.
9.9 Completion Interval
This parameter sets the interval, in seconds, between successive
transmissions of the Completion message during the product
transfer/completion phase.
9.10 Transmit Rate
This parameter specifies the bit rate at which a data product will be
transmitted by the Server. The transmit rate may fall below this
value due to other processing demands on the Server, but the rate
shall never exceed the configured value. The transmit rate applies to
all bits sent by the Server in a DTU, including all protocol headers.
The algorithm to be used to control the transmit rate is unspecified.
9.11 Register Retry Timer
This parameter specifies the frequency, in seconds, that the Client
will resend a Registration message once data transfer phase begins.
During the Announce phase, the Registration message is resent every
time that an Announce message is received that does not confirm the
previous Registration transmission. Hence, the stimulus for
Registration retransmission is the received Announce message and the
rate of retransmission is identical to that of the Announce message
(see Announce Interval parameter above).
9.12 Register Retry Count
This parameter specifies the maximum number of times that the
Client will retransmit a Registration message during the data
delivery phase in order to solicit a confirmation from the Server.
Refer to Register Retry Timer above to see how it is used.
Miller, et. al. [Page 18]
INTERNET-DRAFT MFTP Specification January, 1997
9.13 Response Back-off Timer
This parameter specifies the maximum length of the back-off timer when
suppression is invoked.
9.14 Response TTL
This parameter specifies the TTL of the response message from the
client. It should normally be set to the maximum for the network.
10. Multicast Control Protocol
This section describes the Multicast Control Protocol (MCP).
Configuration parameters and message formats are described in
separate sections.
The purpose of MCP is to allow a Server to control the dynamic
joining and leaving of remote hosts to multicast groups. MCP defines
the following message types:
Join Group transmitted by the Server to dynamically join Clients
to multicast groups. This message is normally unicast
to an individual Client, but may be multicast to an
existing multicast group.
Group Query transmitted by the Client to query a Server for its
current public multicast group address. This message
is unicast to the Server.
Leave Group transmitted by the Server to dynamically cause Clients
to leave a multicast group. This message may be
unicast to an individual Client or multicast to an
existing multicast group.
Echo transmitted by any MFTP station to determine if a
remote host is reachable and running MFTP. This
message may be unicast to an individual host, or
multicast to an existing multicast group.
Echo Response transmitted as a response to a received Echo message.
This message is unicast to the originating host.
MCP provides for creating multicast groups, dissolving multicast
groups, and determining group membership. Each of these is
discussed below. Also refer to the Message Format section for
additional information.
10.1 Creating Multicast Groups (Join Group and Group Query)
Multicast group formation may be initiated by both the Server and
the Client. If a Client knows the IP address of a Server, but not it's
public address where it is announcing product transmissions, it may
send the Group Query message to the Server. The Server then sends
Miller, et. al. [Page 19]
INTERNET-DRAFT MFTP Specification January, 1997
a Join Group message to the Client to get it to join the Server's
public multicast address.
The Join Group message may be sent by a Server at any time to
dynamically direct one or more Clients to join a specified multicast
group. If the Client(s) are not already a member of a multicast group,
the message is sent in unicast or broadcast mode. Otherwise, the
message may be sent to an existing multicast group to direct Clients
joined to that group to join an additional group.
The Join Group message identifies the application type that is sending
the message and the type of data that will be sent by the application.
If the Client has an application layer user of the same type that
wants to receive the indicated data products from this Server, it
joins the multicast group.
There is no response to the Join Group message. However, the Echo
message (see below) may be used to determine which Clients have
successfully joined the group. Since the Echo message can be sent to
the new group address, only those Clients that have actually joined
the new group will respond to the Echo message.
10.2 Dissolving Multicast Groups (Leave Group)
The Leave Group message is used to direct one or more Clients to
leave a specified multicast group. The message may be sent to the
existing multicast group to direct its members to leave the group, or
the message may be sent in unicast or broadcast modes.
There is no response to the Leave Group message. However, the Echo
message (see below) may be used to determine which Clients have
successfully left the group. If the Echo message is sent to the group
address, there should be no response from those Clients that have
left the group.
10.3 Determining Group Membership (Echo and Echo Response)
The Echo message may be sent at any time to determine which
Clients have joined a multicast group. The receiving host responds
with the Echo Response message. The response is sent if MFTP is
running on the target host, without regard to whether there are any
application layer users of MFTP. When users do exist, the receipt of
the Echo message and the transmission of the Echo Response in no
way affects those users.
The Echo message may be multicast, broadcast, or unicast. If the
message is unicast, then the receiving host always responds to the
message. If the message is multicast or broadcast, it may contain a
list of the hosts that should respond to the message. In this case,
the receiving host responds only if it is specified in the host list.
If there is no host list, the receiving host always responds.
The Echo message may be sent by either a Server or Client. It is also
Miller, et. al. [Page 20]
INTERNET-DRAFT MFTP Specification January, 1997
useful for network testing, for example, determine if hosts are
reachable, or to calculate response times.
11. Multicast Data Protocol
This section describes the Multicast Data Protocol (MDP).
Configuration parameters and message formats are described in separate
sections.
The purpose of MDP is to transfer the product data from the Server
to the Clients in a reliable and efficient manner. MDP defines the
following message types:
Announce transmitted by the Server to announce an
impending product transfer. This message is
sent to the Public multicast address.
Registration transmitted by the Client as a response to
the Announce message. This message is sent
to the "Response IP Address" that is
specified in the Announce message
Data Transfer Unit(DTU) transmitted by the Server to deliver a data
product. There is no explicit response to
this message. This message is sent to the
Private multicast address.
Status Request transmitted by the Server to solicit status
from Clients. This message is sent to the
Private multicast address.
Status Response/ACK transmitted by the Client in response to a
Status Request message to acknowledge
correct reception of a block of data
messages. Normally, the Server requests
only the NAK form of the status response to
minimize responses from Clients. This
message is sent to the Response IP Address.
Status Response/NAK transmitted by the Client in response to a
Status Request message to request the
retransmission of data messages within a
block. This is the normal response type
requested by the Server and is the message
that the generic term "status response"
refers to in the following protocol
discussion, unless otherwise noted. This
message is sent to the Response IP Address
(see Definition, Section 2).
Miller, et. al. [Page 21]
INTERNET-DRAFT MFTP Specification January, 1997
Done transmitted by the Client when it has
received all of a product transfer
successfully. This message is sent to the
Response IP Address.
Completion transmitted by the Server to confirm receipt
of a Done message from one or more Clients.
This message is sent to the Private
multicast address.
Abort transmitted by the Server to indicate that
one or more Clients should quit
participating in the current product
transfer, or to indicate the premature
termination of the current product transfer
to all Clients. There is no response. This
message is sent to the Private multicast
address.
Quit sent by the Client to indicate it is no
longer participating in the current product
transfer. There is no response. This message
is sent to the Response IP Address.
MDP consists of three general phases - Product Announcement,
Product Delivery, and Product Completion. Each of these is discussed
below. Also refer to the Configuration and Message Format sections
for additional information.
11.1 Product Announcement
Product Announcement is the process of letting the Client population
know that a product is about to be transmitted. The Server transmits
the Announce message for this purpose. If multicast is being used, the
message is sent on the public "well known" address. Otherwise, the
message is sent in unicast or broadcast mode. The message identifies
the product name, type, and other information about the transfer,
including whether it is Closed, Open Limited or Open Unlimited (see
Group Management above).
The Announce message is sent periodically during the Announce phase.
The transmission interval is specified by the Announce Interval
parameter, and the duration of the Announce phase is specified by the
Announce Time Limit parameter.
The Announce phase serves several purposes. First, it helps to ensure
that the message is actually received by the Clients since it is
transmitted more than once. Second, subsequent transmissions of the
message serve to confirm receipt of Registration messages received
from Clients by the setting of a flag in the message. Third, the
announce duration allows Clients that are experiencing problems (e.g.
insufficient disk space) to resolve those problems and still register
for the product transfer.
Miller, et. al. [Page 22]
INTERNET-DRAFT MFTP Specification January, 1997
When the first Announce message is sent, the Server starts the
Announce Time Limit timer and the Announce Interval timer. When the
interval timer expires, the Server resends the message unless one of
the following conditions is true:
1. The Group type is Closed and all Clients in the Host List (see
below) have registered. Actually, this behavior can be modified
with the Fast Announce Option (see Configuration section). Note
that after the last Client registers, there must be at least one
additional transmission of the Announce message to confirm the
registration. Also, the Server may introduce a delay to ensure that
the last Client has actually joined the Private address before the
data transfer phase is started. In any case, data missed by such
Clients is recovered by subsequent data transmissions of the
Server. No delay is currently specified by this protocol.
2. The Announce type is Open Limited and the maximum number of Clients
that can be handled have registered.
3. The Announce Time Limit timer has expired. If no Clients have
registered when this timer expires, then the product transmission
is canceled.
The Announce message may contain a Host List. It is included for a
Closed Group in order to identify those Clients that are directed to
participate in the product transfer. As Client Registration messages
are received, a flag is set in subsequent transmissions of the
Announce message to confirm receipt of the Registration for each
Client. The Host List is not initially included in the Announce
message for an Open Limited group. However, it is included in
subsequent messages to confirm those Clients that have registered. The
Host List is never included in the Announce message for an Open
Unlimited group.
Clients specified in a Closed group announcement respond with a
Registration message whether they are going to participate in the
transfer or not. If the Client is not going to participate, a reason
is provided in the message. This allows the Server to track problems
in the group.
A Server may receive more than one response from a Client during the
same product announcement. The last response received should be
considered the Client's "official" response. For example, the Client
may initially decline the transfer due to insufficient disk space. The
announce duration may allow the Client operator time to resolve the
problem. If the Client operator frees up disk space, the Client may
then respond that it will be participating in the transfer.
The Server may receive a Quit message from the Client after the Client
has registered. After receipt of the sent message, the Server should
not consider the Client a participant in the product transfer.
The logic flow diagram below illustrates the Server's product
Miller, et. al. [Page 23]
INTERNET-DRAFT MFTP Specification January, 1997
+-+-+-+-+-+-+-+-+
| Idle |<---------------------------+
User request to +-+-+-+-+-+-+-+-+ |
transmit prod. --------------->V |
+-+-+-+-+-+-+-+-+ |
| Start | |
|duration timer | |
+-+-+-+-+-+-+-+-+ |
V |
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+ |
| Send |<------| Update | |
| Announce | | confirmations | |
+-+-+-+-+-+-+-+-+ +-+-+-+-+-+-+-+-+ |
V ^ |
+-+-+-+-+-+-+-+-+ | |
| Start | | |
|interval timer | | |
+-+-+-+-+-+-+-+-+ | |
V | |
+-+-+-+-+-+-+-+-+ | |
Yes | Registration |<-----+ | |
+---------| fulfilled? | | | |
| +-+-+-+-+-+-+-+-+ | | |
| V No | | |
| +-+-+-+-+-+-+-+-+ | | |
| Yes | Duration timer| | | |
| +----| expired? | | | |
| | +-+-+-+-+-+-+-+-+ | | |
| | V No | | |
| | +-+-+-+-+-+-+-+-+ | | |
| | |Interval timer | No | | |
| | | expired? |+-----+ | |
| | +-+-+-+-+-+-+-+-+ | |
| | | Yes | |
| | +-----------------------+ |
| | +-+-+-+-+-+-+-+-+ |
| | | Anyone | No |
| +--->| register? |+---------------------------+
| +-+-+-+-+-+-+-+-+
| V Yes
| +-+-+-+-+-+-+-+-+
| | Send |
+-------->|final Announce |
+-+-+-+-+-+-+-+-+
V
+-+-+-+-+-+-+-+-+
| Start data |
| delivery |
+-+-+-+-+-+-+-+-+
Figure 11-1
Server Product Announcement Phase
Miller, et. al. [Page 24]
INTERNET-DRAFT MFTP Specification January, 1997
announcement phase. "Registration fulfilled?" refers to whether all
Clients in a Closed group have registered, or whether the maximum
number of Clients in an Open Limited group have registered (it does
not apply at all to an Open Unlimited group). "Update confirmations"
refers to the setting of the registration-confirmed flag in the
Announce message to confirm Client registrations. This step also would
not apply to Open Unlimited groups.
When a Client receives an Announce message, it examines the product
information contained in the message and compares it with the
requirements set by its application layer user(s). That is, it is
expected that the application layer will designate the Servers that it
is interested in receiving from and the types of data products that it
wants to receive. Based on this information, the Client decides
whether or not to receive the announced product.
If the Announcement is for Open Limited or Open Unlimited Groups and
the Client does not want to receive the product, then the Client sends
no response and discards the message. If the Announcement is for
Closed Groups and the Client does not want to, or is not able to
receive the product, then it still must send a Registration message
stating the reason it is declining the transfer.
If a Client wants to participate in a transfer, its action is
described separately for each group management type below. Note that
joining and leaving the multicast group in the discussion below refers
to IGMP operations and not MCP.
11.1.1 Closed Group
If the Client's IP address is not contained in the Host List of the
Announce message, the Client sends no message and is administratively
prohibited from participating in the transfer. Otherwise, it joins the
data delivery multicast address (i.e. Private address) contained in
the Announce message and sends a Registration message to the Response
Address.
The Client's registration must be confirmed by the Server before it is
allowed to receive the data product. The confirmation is indicated by
a flag set in a subsequent Announce message, as discussed above. If
the confirmation is received, the Client waits for and receives the
product data. If confirmation is not received in the next Announce,
the Client resends its Registration message.
If the announce duration (specified in Announce message, see Messages
below), expires or the Client begins to receive product data before
confirmation is received, it implies that the Server did not receive
the Registration message before the start of the data delivery phase.
The Client discards the received data and resends its Registration
message at the Register Retry Timer interval. The number of times that
the Client resends its Registration message is limited by the Register
Retry Count. If no confirmation is received before the retry limit is
Miller, et. al. [Page 25]
INTERNET-DRAFT MFTP Specification January, 1997
reached, the Client returns to idle state and makes no further
attempts to register for the current product transfer. Note that the
Client also returns to idle state if a Completion or Abort message is
received while the Client is still trying to register with the Server.
11.1.2 Open Limited Group
The Client joins the data delivery address contained in the Announce
message and sends a Registration message to the Response Address.
Subsequent operation is identical to Closed Groups in that the Client
must receive confirmation of its registration before it is allowed to
receive the product data.
11.1.3 Open Unlimited Group
The Client joins the data delivery address contained in the Announce
message but sends no Registration message to the Server. The Client
receives the data as sent by the Server on the data delivery address.
A Client may receive an Abort message from the Server at any time
during the product registration. The Client returns to idle state and
is not allowed to participate in the current product transfer. When
returning to idle state, the Client normally leaves the data delivery
group.
However, the Client would remain a member of the group if:
1. The Server specified in the Announce message that Clients
should not leave the data delivery group (see Announce
message below).
2. Other transmissions being received by this Client are using
the same data delivery group address.
3. The private address is the same as the public address.
11.2 Product Delivery
The Product Delivery phase occurs after the Product Announcement
phase. The file product is transmitted by the Server to the Clients.
11.2.1 File Delivery - Server
DTUs are transmitted to the private multicast address. The Server
logically divides the file into one or more parts, called blocks. Each
block is further divided into the data size that can be transmitted in
a single DTU. All blocks and DTUs are a fixed size, except the last
DTU which may be shorter. The manner in which the block size is
determined is discussed further below. This data organization is
illustrated in Figure 11-2.
Miller, et. al. [Page 26]
INTERNET-DRAFT MFTP Specification January, 1997
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| File |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| |
| |
| |
+-+-+-+-+-+-+-+-+-+-+ +-+-+-+-+-+
| Block 1 | Block 2 | --- | Block n |
+-+-+-+-+-+-+-+-+-+-+ +-+-+-+-+-+
/ \
/ \
/ \
/ \
/ \
/ \
/ \
+-+-+-+-+-+-+ +-+-+-+
|DTU1 |DTU2 | --- |DTUn |
+-+-+-+-+-+-+ +-+-+-+
Figure 11-2
Organization of Transmitted Data
The Server transmits the file data in "passes". That is, the entire
file is sent initially (i.e. pass 1). If any DTU retransmissions are
required, the Server then makes another pass (i.e. pass 2) through the
file, but sends only those DTUs that were reported as missed by the
Clients. Additional passes may be required to successfully deliver all
DTUs to all Clients.
The pass, block, and DTU number of each file fragment is indicated in
the DTU header (see Message Formats below). The block and DTU number
assignment is static for a given MTU size such that the retransmission
of any file fragment will indicate the same block and DTU number. This
allows the receiver to properly insert each fragment into the file
regardless of when the fragment is received.
The Server uses the Status Request message to query the Clients for
their receive status. The Clients send a Status Response message if
they need to request the retransmission of any DTUs. Otherwise, a
Client sends no Status Response to the request. Although Client status
may be requested by the Server at any time, it is normally requested
at the end of each block transmission. The response message contains a
bit map where each bit represents the receive status of an individual
DTU within the block. Hence, the status of an entire block is
contained within a single response message.
The Server does not stop at the end of each block and wait for a
response to the Status Request message nor does it immediately resend
requested DTUs. Rather, the Server simply receives and processes the
responses in order to schedule which DTUs need to be resent on the
next pass. This makes the protocol insensitive (unless the data
Miller, et. al. [Page 27]
INTERNET-DRAFT MFTP Specification January, 1997
product sent is very short) to network delay that can be excessive in
some networks, e.g. satellite. Note that the Status Request message
may be sent at any time rather than at block boundaries, if
information is desired about transmissions before the end of the
block.
After the last block of a pass has been transmitted, the Server sends
a Status Request message for that block and immediately starts the
next pass if there are DTUs that need to be resent. However, if no
responses have been received for any previous block in the file, the
Server stops and waits for responses. The length of time that the
Server waits for a response is specified by the Status Delay parameter
(see Configuration above). Note that the Status Request message may
specify that status is requested for a single block or for a range of
blocks. When the Server sends the Status Request message because no
responses have been received, it specifies all blocks in the file. If
a Client has not responded because it missed a previous individual
block message, this ensures that such responses are solicited.
Each Status Request message specifies the current transmit pass number
and the block for which status is being requested. These values are
echoed by the Client in its response. The Server uses the following
rules for processing Client responses.
1.If the response pass number is equal to the current transmit
pass number, accept the response and schedule the requested
DTUs for retransmission.
2.If the response pass number is not equal to the current
transmit pass number, examine the block number and proceed as
follows:
a. If the response block number is greater than the
current transmit block number, accept the response and
schedule the requested DTUs for retransmission.
b. Else, schedule for retransmission only those DTUs that
have not already been retransmitted on this pass.
The block size is a function of the Status Interval and Transmit Rate
parameters. That is, the application layer user specifies the
frequency at which receive status should be obtained from the Clients.
The Server computes how many DTUs it can send at the Transmit Rate in
the Status Interval and establishes that as the block size. The block
size is computed with the following formula:
block size = Status Interval x (Transmit Rate/(8 x (MTU - protocol
headers)))
There is an upper limit to the block size that the application may set
per Transmit Rate, as determined by the maximum DTU bit map size that
the MTU will allow.
Miller, et. al. [Page 28]
INTERNET-DRAFT MFTP Specification January, 1997
The default block size (used when application does not specify a
Status Interval) maximizes the amount of status information that can
be returned in a single message according to the MTU size.
Retransmissions may continue until all Clients have received the
entire file or until the Delivery Time Limit duration has expired or
until the Status Retry Limit has been exceeded. If the timer expires,
the Server sends an Abort message and terminates the product delivery
unless it is an Open Unlimited group, at which time a Completion
message is sent. It is required to send several Abort messages to
ensure that the Client receives at least one message since there is no
confirmation message from the Client.
The logic flow diagram in Figure 11-3 illustrates the Server's Product
Delivery phase. The completion processing is explained in a following
section.
If a file is being restarted by the Server, the Server must first
determine what part of the file has not yet been received by the
Clients. There are many possible ways this can be done and no specific
way is specified by the protocol. For example, the Server may locally
store the state at the end of an aborted transfer and restore it for a
file restart. Alternatively, the Server may want to query one or more
Clients to determine what parts of the file still need to be
transmitted. Once this initial restart state is determined, subsequent
Server operation is identical to any retransmission pass (i.e. the
Server sends only those DTUs necessary to complete the file transfer).
File restart is allowed for Closed and Open Limited groups, but not
for Open Unlimited groups.
11.2.2 File Delivery - Client
As DTUs are received by the Client, the Client must examine the block
and DTU number in order to store the data in its proper place in the
file. As discussed above, a bit map of each block is maintained that
indicates whether a DTU has yet been received. If a duplicate DTU is
received, it is discarded.
When a Status Request message is received from the Server, the Client
immediately responds with a Status Response message for the indicated
block. File delivery continues until all DTUs in the file have been
received or until the receive timer expires. The receive timer value
is specified by the Server in the Announce message. If the timer
expires, the Client sends a Quit message and returns to idle state. It
is required to send several Quit messages to ensure that the Server
receives at least one message since there is no confirmation message
from the Server.
Miller, et. al. [Page 29]
INTERNET-DRAFT MFTP Specification January, 1997
from Announce +-+-+-+-+-+
phase | Idle |
| +-+-+-+-+-+
| ^
V |
+-+-+-+-+-+-+-+ +-+-+-+-+-+
+--->|Xmit duration| Yes | Send |
| | expired? |----->| Abort |
| +-+-+-+-+-+-+-+ +-+-+-+-+-+
| |No
| V
| +-+-+-+-+-+-+-+
| | Send |
| | DTU |
| +-+-+-+-+-+-+-+
| |
| V
| +-+-+-+-+-+-+-+
|No | End of |
|<---| block? |
| +-+-+-+-+-+-+-+
| | Yes
| V
| +-+-+-+-+-+-+-+
| | Send Status |
| | Request |
| +-+-+-+-+-+-+-+
| |
| V
| +-+-+-+-+-+-+-+
|No | End of |
|<---| pass? |
| +-+-+-+-+-+-+-+
| | Yes
| V
| +-+-+-+-+-+-+-+
|Yes | Retrans. |No Completion
+<---| required? |----> processing
+-+-+-+-+-+-+-+
Figure 11-3 Server Product Delivery
The logic flow diagram in Figure 11-4 illustrates the Client's receive
data processing. The completion processing is explained in a following
section.
11.3 Product Completion
Product Completion is the process of terminating a product transfer.
File product completion for the Client is discussed first, since it is
the Client that normally initiates the completion procedure.
Miller, et. al. [Page 30]
INTERNET-DRAFT MFTP Specification January, 1997
from Announce +-+-+-+-+-+
phase | Idle |
| +-+-+-+-+-+
| ^
V |
+-+-+-+-+-+-+-+ +-+-+-+-+-+
+--->|Receive timer| Yes | Send |
| | expired? |----->| Quit |
| +-+-+-+-+-+-+-+ +-+-+-+-+-+
| |No
| V
| +-+-+-+-+-+-+-+
| | Receive |
| | DTU |
| +-+-+-+-+-+-+-+
| |
| V
| +-+-+-+-+-+-+-+
|No | Received |
|<---| Status Req.?|
| +-+-+-+-+-+-+-+
| | Yes
| V
| +-+-+-+-+-+-+-+
| | Send Status |
| | Response |
| +-+-+-+-+-+-+-+
| |
| V
| +-+-+-+-+-+-+-+
|Yes | Retrans. |No Completion
+<---| required? |----> processing
+-+-+-+-+-+-+-+
Figure 11-4. Client Product Reception
11.3.1 File Product Completion - Client
As each DTU is received, the Client determines if it has now received
the entire file. If not, data reception continues. Otherwise, for
Closed and Open Limited groups, the Client sends a Done message to the
Server. For Open groups, the Client simply returns to idle state
without sending any message to the Server.
After sending a Done message, the Client waits for receipt of a
Completion message from the Server. Like the Announce message, the
Completion message contains a Host List that specifies each Client
Miller, et. al. [Page 31]
INTERNET-DRAFT MFTP Specification January, 1997
for which a Done message has been received (i.e. it confirms receipt
of the Done message).
If the Client receives a Completion message and it's address is in the
Host List, then the transfer is complete, the Client leaves the data
delivery group (subject to conditions, as discussed above) and returns
to idle state. Otherwise, the Done message is resent. The Done message
is also resent for each Status Request message received from the
Server (a Status Response is not sent). This procedure continues until
the Client's Done message is confirmed or until the receive timer
expires.
The logic flow diagram below illustrates the Client's completion
processing as it applies to a Closed or Open Limited group.
from Delivery
phase
|
V
+-+-+-+-+-+-+-+
+--->| Send |
| | Done msg. |
| +-+-+-+-+-+-+-+
| |
| V
| +-+-+-+-+-+-+-+ +-+-+-+-+-+
| | Done | Yes | Idle |
+-------->| confirmed? |----->| |
| | +-+-+-+-+-+-+-+ +-+-+-+-+-+
| | |No ^
| | V |
| | +-+-+-+-+-+-+-+ |
| |Yes | Status Req. | |
| +<---| received? | |
| +-+-+-+-+-+-+-+ |
| |No |
| V |
| +-+-+-+-+-+-+-+ +-+-+-+-+-+
| No |Receive timer| Yes | Send |
+<--------| expired? |----->| Quit |
+-+-+-+-+-+-+-+ +-+-+-+-+-+
Figure 11-5 Client Product Completion
11.3.2 File Product Completion - Server
For Closed and Open Limited groups, the receipt of a Done message from
a Client means that the Client has received all of the file
successfully. The Server confirms the receipt of Done messages with
the Completion message. However, a separate message is not sent for
each Done received. Rather, the Host List is used to "batch-confirm"
all Done messages that have been received up to the time that the
Completion message is sent.
Miller, et. al. [Page 32]
INTERNET-DRAFT MFTP Specification January, 1997
The first Completion message is sent as soon as the first Done message
is received, and periodically thereafter at intervals specified by the
Completion Interval parameter. When a Completion message is sent, the
interval timer is started. When the timer expires, the Server updates
the Host List of the message with all Clients that have sent Done
messages and sends the message again. If all Clients have not yet
completed, the timer is restarted. There must be at least one
additional transmission of the Completion message after the last
Client has completed.
Since Clients will experience different network error rates, some
Clients may complete the transfer while other Clients are still
receiving retransmissions. This means that the Completion message may
be sent while the file transfer is still in progress.
Data delivery continues until there are no longer any participating
Clients (all Clients have sent Done or Quit messages), there is no
response from any Client, or until the Delivery Time Limit duration
has expired, whichever occurs first.
When the last Client completion has been confirmed with the Completion
message, the Server considers the transfer complete and sends no
further messages.
If the Server receives no response to a Status Request sent at the end
of a pass, there may no longer be any Clients listening to the Server.
The Status Delay and Status Retry Count parameters limit the length of
time that the Server will attempt to receive a response. When the
Status Request message is sent, the Server starts the Status Delay
timer. If any status responses have been received by the time the
delay timer expires, the Server begins the next pass where it
retransmits the DTUs specified in the responses. If no responses are
received, the Server sends another Status Request message and restarts
the delay timer. This procedure repeats up to the limit set by the
Status Retry Count parameter, or the delivery time limit expires,
whichever occurs first. At that time, the Server sends an Abort
message, terminates the product delivery, and sends four End
messages over the Public Address. End messages are read by
aggregation devices, if present, to determine that the session is
over.
The Server may also use the Abort message to terminate transfer to one
or more Clients at any time due to conditions at the Server or at the
Client. Several Abort messages are sent to insure delivery as no
response to the Abort message is required from the Clients.
For Open Unlimited groups, the Clients do not send Done messages. The
Server simply continues to send retransmissions as long as there are
responses to the Status Request message. When there are no responses,
the Server continues to send Status Request messages up to the Status
Retry Limit (as discussed above) or the Delivery Time Limit expires.
At that time, the Server sends a Completion message (with no Host
List) and considers the transfer complete. There is no response from
Miller, et. al. [Page 33]
INTERNET-DRAFT MFTP Specification January, 1997
the Clients.
For all group types, when the session is over the Server sends an End
message on the same multicast address used for Announce. This message
is sent four times and is used to conveniently notify infrastructure
that may be aggregating that the session is ended. The End message
structure is described in Section 12.
11.4 Flow Control
A first level of flow control may optionally be invoked to make sure
particular parts of the network multicast tree do not become
overloaded and bad receivers do not hold up transmission to the other
members of the group.
The Announce message may optionally contain an error rate threshold
(Note that this should probably be mandatory when MFTP is used in
large heterogeneous networks such as the Internet). This threshold is
used to inform clients that if they measure a DTU loss over the
threshold they abort themselves from the group. By leaving the group,
that leaf of the multicast tree is dissolved, removing congestion in
that part of the network. The Server may then be directed by the
application to set up a new group of aborted clients and create a new
session at a later time at a lower transmission rate to service those
clients.
12. Message Formats
This section defines the format of all MFTP protocol messages. All
messages begin with a common protocol header that is 12 bytes in
length and includes the message type. Additional information,
depending on message type, is included in the message payload area.
The format of the common header is shown in Figure 12-1 below.
0 32
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---
| Protocol version | Message type |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Common
| Message length | Checksum |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ Header
| Transmission Identifier |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---
| Optional Parameters and Data |
| |Payload
~ ~ Area
| |
| |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ ---
Figure 12-1 Protocol Header Format
Miller, et. al. [Page 34]
INTERNET-DRAFT MFTP Specification January, 1997
Each header field is an unsigned integer. The fields are described
below.
Protocol Version Number
This field contains an integer that identifies the MFTP protocol
version that is implemented by the sender of this message. This
document describes version 1.
Message Type
This field identifies the MFTP message type, as defined in the
following table.
QUERY GROUP 0x0001
JOIN GROUP 0x0002
LEAVE GROUP 0x0003
ANNOUNCE 0x0004
DATA TRANSFER 0x0005
STATUS REQUEST 0x0006
NAK RESPONSE 0x0007
COMPLETION 0x0008
ABORT 0x0009
REGISTRATION 0x000A
DONE 0x000B
QUIT 0x000C
ECHO 0x000D
ECHO RESPONSE 0x000E
ACK RESPONSE 0x000F
END 0x0011
Message Length
This field contains the length, in bytes, of the entire MFTP message,
beginning with the Protocol Version field.
Checksum
This field contains a checksum that verifies the integrity of the
entire message including the MFTP header. A checksum is computed by
the sender of each message and placed into this field. The receiver
also computes the checksum and compares it with the received value. If
the values do not match, the received message is discarded and treated
as if it were never received.
The checksum is the 16 bit one's complement of the one's complement
sum of all 16 bit words in the header and payload area. While
computing the checksum, the checksum field itself is replaced with
zeros.
The use of this field is optional as the UDP checksum provided within
UDP should be used for this function. If the UDP checksum is used,
this field should be set to zero.
Miller, et. al. [Page 35]
INTERNET-DRAFT MFTP Specification January, 1997
Transmission Identifier
This field contains a number that uniquely identifies this
transmission. A receiver uniquely identifies messages associated with
this transmission by examining the source IP address and this
transmission identifier. The transmission identifier is not
applicable in MCP and is set to zero in MCP messages.
Payload
The information contained in the payload section of each message is
parameterized. This is done for several reasons. First, it facilitates
the addition of new parameter types as the protocol evolves. Second,
it allows for the omission of parameters that do not pertain to a
particular product type. This is particularly important in the
Announce message.
All parameter data fields are a multiple of four bytes (i.e. 32 bits).
When such a field is used to contain an integer, its format is as
shown below. The integer is represented in two's complement notation.
The most and least significant bytes are 0 and 3, respectively.
MSB LSB
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Byte 0 | Byte 1 | Byte 2 | Byte 3 |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
<--------------------------- 32 Bits --------------------------->
Figure 12-2 Basic Field Format
The type-length-value structure is shown below. The minimum parameter
unit length is 64 bits. The Name and Length fields both consist of
two 16 bit fields. The value field is variable length, but is always
a multiple of 32 bits. When the value field contains a character or
octet string that is not a multiple of 32 bits, the field is extended
so that the 32 bit multiple requirement is always met. This means that
1, 2, or 3 pad bytes are added to the end of the string, and the value
of those bytes is set to zero. However, the length field defines the
true length of the value.
|<------------ 32 bits -------->| Variable
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - - -+
| Type/length | Value |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ - - -+
Figure 12-3 Type-Length-Value Format
Each item of data that is conveyed in the payload area of the message
is identified as a separate message parameter and structured with the
type-length-value format. Each parameter may be used in one or more
message types.
The following table defines the complete list of parameters and
Miller, et. al. [Page 36]
INTERNET-DRAFT MFTP Specification January, 1997
includes the parameter description, type, and length.
Parameter | Type | Data Length
---------------------------------------------------------------
Announce Duration 0x0100 4
Announce Options 0x0101 4
Application Reference 0x0102 variable(1-64)
Block Number 0x0103 4
Block Range 0x0104 8
Block Size 0x0105 4
Cancel Reason 0x0106 4
Client Address 0x0107 4
Data Rate 0x0108 4
Data Transfer Duration 0x0109 4
DTU Number 0x010A 4
DTU Range 0x010B 8
DTU Size 0x010C 4
Echo Sequence 0x010D 4
Echo Data 0x010E variable(1-4096)
Error Threshold 0x010F 4
Event Count 0x0110 4
Event Hash 0x0111 16
Event List 0x0112 variable
Host List(complex) 0x0113 variable
Host List(simple) 0x0114 variable
Group Options 0x0115 4
Multicast Address 0x0116 4
Product Data 0x0117 variable
Product Name 0x0118 variable (1-256)
Product Size 0x0119 4
Private Address 0x011A 4
Public Address 0x011B 4
Registration Response 0x011C 4
Response Address 0x011D 4
Response Port 0x011E 4
Serial Number 0x011F 4
Server Address 0x0120 4
Server Port 0x0121 4
Status Type 0x0122 4
User Data 0x0123 variable (1-64)
Figure 12-4 Message Parameters
Each parameter is described in the message definition sections below
so that their use is understood in the context of each message type.
Parameters may be required, optional, or omitted depending on the
message type and data product type.
12.1 MCP Messages
The MCP messages have been listed in the Protocol Description section
above. The format of each message is described below. All messages
begin with the common protocol header (see Protocol Header above).
Miller, et. al. [Page 37]
INTERNET-DRAFT MFTP Specification January, 1997
All subsequent fields are used to convey parameters to the receiving
host, and use the type-length-value format. Each message description
provides a list of the parameters included in that message type and
whether each parameter is required or optional. The parameters may
appear in the message in any order, not necessarily in the order
presented in the list.
12.1.1 Query Group Message
The purpose of the Query Group message is to allow a Client to
determine a Server's public address when only the Server's IP address
is known. The message is sent in unicast mode to the Server. The
response to this message is a Join Group message, also sent in unicast
mode. A separate Join Group is returned for each public address active
at the Server. If there are no public addresses active, then no
response is sent by the Server. The Query Group message contains the
following parameters.
Application Reference String (Mandatory)
Group Options (Mandatory)
Application Reference String
Given that there may be different (i.e. incompatible) application
layer users of MFTP, this parameter serves to identify the application
type operating at the Server and the Clients.
The parameter is an ASCII character string that identifies the
application type. For example, the StarBurst file transfer application
uses the string "StarBurstMFTP".
In a Client message, such as Query Group, the application running at
the Client is identified. Hence, the Server responds to the message
only when an application of the same type is running at the Server. In
a Server message, such as Join Group (see below), this parameter
identifies the application type that is running at the Server. The
Client may then use the information to determine if it will respond to
the Server message. That is, only Clients that have the same type
application layer users should respond to the message.
Group Options
This parameter defines various options associated with group
management and is contained in both the Group Query and Join Group
messages. The parameter is a combination bit and byte field, as
defined below:
Miller, et. al. [Page 38]
INTERNET-DRAFT MFTP Specification January, 1997
Byte Bit Setting
PRODUCT TYPE
0 0-7 'F' = file product
'A' = all product types
COMMON OPTIONS
1 0-7 Unused, set to zero
2 0-7 Unused, set to zero
3 0-7 Unused, set to zero
Each option is explained below:
Product Type This option identifies the product type(s) that a
Client wants to receive (as used in Query Group
message) or that a Server is transmitting(as used
in Join Group message). Only file product is now
defined.
Common Options None currently defined.
12.1.2 Join Group Message
This message may be sent as a response to the Query Group message (see
above), or as an unsolicited message to one or more Clients. In the
former case, it provides the Server's public multicast address to the
Client so that the Client may join the group to receive product
announcements. In the latter case, it may specify the public address,
or any other multicast address that the Server wants the Client(s) to
join. When sent as a response to the Query Group message, it is sent
in unicast mode. Otherwise, the message may be sent in unicast,
broadcast, or multicast modes. The message contains the following
parameters.
Application Reference (mandatory)
Multicast Address (mandatory)
Group Options (mandatory)
Host List (simple) (optional)
Application Reference String
Refer to the Query Group message for a description of this parameter.
The Client joins the indicated group only if the Client has an
application user of the same type indicated in the message.
Multicast Address
This parameter contains the multicast address that the receiving host
should join. Normally, this will be the Server's public address, but
may be any multicast address that the Server wants the Client(s) to
join.
Group Options
Refer to the Query Group message for a description of this parameter.
Miller, et. al. [Page 39]
INTERNET-DRAFT MFTP Specification January, 1997
The Server sets the Product Type field of this parameter to identify
the type of data products that may be received by joining the
indicated public multicast address.
Host List (simple)
This parameter is included when the sender wants to specify specific
hosts that should act on the message. If the message is to apply to
all receiving hosts, this parameter is omitted.
This parameter is an array of host records. Each host record in the
array consists of a single 32 bit field, called the host identifier.
Each host identifier contains the IP address of a host. The address is
stored in network byte order.
There are other messages that may contain the Host List parameter
(e.g. Announce, Abort). Whenever the length of the Host List does not
fit within a single message, the Host List is divided and sent in two
or more copies of the same message type. The multiple messages are
considered a single logical message by the Server.
12.1.3 Leave Group Message
The purpose of the Leave Group message is to direct one or more remote
hosts to leave a multicast group. The message may be sent in unicast,
broadcast, or multicast modes. The message contains the following
parameters.
Multicast Address (mandatory)
Host List(simple) (optional)
Multicast Address
This parameter specifies the multicast group that the Client(s) should
leave.
Host List (simple)
Refer to the Join Group message for a description of this parameter.
12.1.4 Echo Message
The Echo message may be transmitted by any MFTP station. The purpose
of the message is to test the network path, and the reliable transfer
of data over that path, between the source and destination hosts. The
response to this message is the Echo Response message (see next
section). This message is similar in concept to the Echo message of
the Internet Control Message Protocol (RFC792). This message may be
sent in unicast, broadcast, or multicast modes. The message contains
the following parameters.
Miller, et. al. [Page 40]
INTERNET-DRAFT MFTP Specification January, 1997
Echo Sequence (mandatory)
Echo Data (optional)
Host List(simple) (optional)
Echo Sequence
This parameter is a sequence number to aid in matching a response with
a transmitted message.
Echo Data
This optional parameter is an octet string that can be sent with the
Echo message and is returned in the Echo Response message.
Host List (simple)
Refer to the Join Group message for a description of this parameter.
12.1.5 Echo Response Message
The Echo Response message may be transmitted by either a Server or
Client, as a response to a received Echo message. This message is sent
in unicast mode always.
All parameters received in the Echo message, except the host list, are
echoed back to the sender exactly as received. The host list, if
contained in the Echo message, is omitted from the response. Refer to
the Echo Message above.
12.2 MDP Messages
The MDP messages have been listed in the Protocol Description section
described previously. The format of each message is described below.
All messages begin with the common protocol header (see Protocol
Header, Figure 12-1). All subsequent fields are used to convey
parameters to the receiving host, and use the type-name-length-value
format. Each message description provides a list of the parameters
included in that message type and whether each parameter is required
or optional.
Unless otherwise stated, the parameters in messages sent by the Server
may appear in the message in any order, not necessarily in the order
presented in the list. To allow efficiency in aggregating response
messages (when used), most response formats specify that the
parameters must be in the order shown.
12.2.1 Announce Message
The Announce message is sent only by the Server. The purpose of the
message is to allow the Server to announce the impending transmission
of a data product, to describe the data product, and to solicit
Registration message responses from Clients (depending on group type).
The Announce message may be sent in unicast, broadcast, or multicast
Miller, et. al. [Page 41]
INTERNET-DRAFT MFTP Specification January, 1997
mode. In multicast mode, it is sent to the Public address.
The Announce message contains the parameters listed below. An
intermediate network entity that is performing response aggregation
may need to monitor Announce messages in order to obtain parameters
that are needed for aggregation processing, such as the Response
Address. Therefore, the first five parameters (Server Address/Port,
Response Address/Port, and Private Address) must appear at the
beginning of the message payload area and in the order shown. Other
parameters may appear in any order.
Server Address (mandatory)
Server Port (mandatory)
Response Address (mandatory)
Response Port (mandatory)
Private Address (mandatory)
Announce Duration (mandatory)
Announce Options (mandatory)
Application Reference (mandatory)
Block Size (mandatory)
Data Rate (mandatory)
Product Name (mandatory)
Data Transfer Duration (mandatory)
DTU Size (mandatory)
Product Size (mandatory)
Error Threshold (optional)
User Data (optional)
Host List (complex) (optional)
Server Address
This parameter is the IP address, in network byte order, of the
Server. Its purpose is primarily for response aggregation. This
parameter is copied to the response by the Client so that the
aggregating entity knows where to send the aggregated response.
Server Port
This parameter is the receive UDP port number of the Server. Its
purpose is primarily for response aggregation. This parameter is
copied to the response by the Client so that the aggregating entity
knows where to send the aggregated response.
Response Address
This parameter is the address to which Client response messages are to
be sent. This allows responses to be sent to or intercepted by an
intermediate aggregating entity. The address may be a unicast or
multicast address, in network byte order. If response aggregation is
not being performed, this address is the host IP address of the
Server.
Miller, et. al. [Page 42]
INTERNET-DRAFT MFTP Specification January, 1997
Response Port
This parameter is the receive UDP port at the Response Address to
which Client response messages are to be sent.
Private Address
This parameter specifies the network address to which the Server will
transmit the product data. For a multicast transmission, this is a
multicast group address. The Client must join the multicast group to
receive the product data. Otherwise, this parameter specifies a
unicast or broadcast address.
Announce Duration
This parameter is the time, in seconds, remaining before the product
data transmission will begin. It is updated at each Announce message
transmission so that it accurately reflects the remaining time. The
Client does not act on this information. It is provided primarily as
an item of information to the application layer user. The initial
value of this parameter is provided by the Announce Time Limit
configuration parameter (see Configuration above).
Announce Options
This parameter defines various options associated with a product
transmission. The parameter is a combination bit and byte field, as
defined below:
Byte Bit Setting
PRODUCT TYPE
0 0-7 'F' = file product
COMMON OPTIONS
1 0-1 01 = closed group
10 = open limited group
11 = open unlimited group
2 1 = do not leave delivery group
0 = leave delivery group
3-7 Unused, set to zero
FILE OPTIONS
2 0 1 = initial transmission
0 = restart transmission
1 1 = overwrite existing file
0 = preserve existing file
2 1 = late entry allowed
0 = late entry disallowed
3-7 Unused, set to zero
Each option is explained below:
Product Type This option identifies the product type. Only
file product is currently defined.
Miller, et. al. [Page 43]
INTERNET-DRAFT MFTP Specification January, 1997
Common Option 1 This group option defines whether the product
(bit 0-1) may be received by any Client, or only by those
specified in the Host List parameter (refer to
the Group Management section above)
Common Option 2 This option specifies whether the Client should
(bit 2) or should not leave the data delivery group
after the product delivery has completed. If
the Server intends to send more than one
product to the same Client group, then group
setup time can be minimized by keeping Clients
joined to the group. Since this option is
contained in each product announcement, the
final product announcement can direct Clients
to leave the group after the transmission has
completed.
File Option 1 The initial/restart option indicates whether
this is an initial transmission of the file
(i.e. the entire file will be sent) or whether
it is a restart of a file previously sent but
not received completely by all Clients.
File Option 2 The overwrite/preserve option indicates whether
a Client should overwrite or preserve a current
file that has the same name as specified in the
Product Name parameter (see above). If this
option is set to overwrite, then the Client is
allowed to receive the new file and overwrite
the old file. If this option is set to
preserve, then the Client declines to receive
the new file and preserves the existing file at
the host.
File Option 3 The late entry option defines whether a Client
may participate in a file restart if it has not
yet received any of the file. Since the file
will have to be entirely transmitted for the
Client, it will slow down other Clients that
may need only a small portion of the file to
complete their receipt of it. However, it does
allow the Server to transmit the file to
Clients that may have been offline when the
file was originally sent. If the option is set
to allowed, then a Client that has not yet
received any of the file may participate in the
file transfer. If the option is set to
disallowed, then only Clients that have
previously received some portion of the file
may participate.
Miller, et. al. [Page 44]
INTERNET-DRAFT MFTP Specification January, 1997
Application Reference String
Refer to the Join Group message for a description of this parameter.
The Client will not respond to this message unless its application
type matches this value.
Block Size
This parameter is the number of DTUs that will be transmitted in each
block. It allows the Client to initialize and manage its memory
resources used for tracking DTUs that have been received or missed. If
the product is a file, this value is limited by the maximum size of
the bit map form of the Event List that can be carried in a Status
Response message (see Event List below).
Data Rate
This parameter specifies the Server's transmit data rate. The Client
may use this information to decide whether it can successfully
participate in the product transfer. That is, the Client might decline
a product transfer if it knows that the data rate will overrun the
Client.
Data Transfer Duration
This parameter specifies the maximum amount of time, in seconds, that
the Server will devote to the product delivery (not including the
product announcement phase). The Client uses this value as its receive
timeout. This parameter is configured (see Configuration section
above).
DTU Size
This parameter is the maximum number of data bytes that will be
contained in each DTU for a file transfer. The last DTU may contain
less than this number of bytes. This value is computed by MFTP based
on the size of the network Maximum Transmission Unit, the lower level
protocol headers, and the other parameters that must reside in a DTU.
Error Threshold
This parameter defines the percentage of missed DTUs, relative to the
total number of DTUs in the transmission, that can occur at a Client
before the Client voluntarily removes itself from participation in the
product transfer. That is, the Client monitors its own error rate and
initiates its own removal from the transfer if this threshold is
reached. This mechanism helps to prevent a small number of Clients
with high error rates from adversely affecting all other Clients by
requesting a high percentage of retransmissions. It is permissible for
the Client to continue to receive and process DTUs in anticipation of
a file restart, but the Client must not send any message to the
Server.
Miller, et. al. [Page 45]
INTERNET-DRAFT MFTP Specification January, 1997
Product Name
This parameter is an ASCII character string that is the formal name
for the product being announced. If the product is a file, this field
contains the name of the file that should be created on the Client
system to store the product (the filename may be different than the
filename at the Server).
Product Size
This parameter is the total length, in bytes, of a file product. It
allows the Client to determine if there is sufficient disk space to
receive and store the product.
User Data
This parameter is an octet string that is provided by the application
layer user. It is delivered to the application layer user at the
Client. It could be used, for example, to describe the product being
announced. Its use is optional and is limited to 64 octets.
Host List (complex)
This parameter is an array of host records. Each host record in the
array consists of two 32 bit fields. The first field, called the host
identifier, contains the IP address of the remote host, in network
byte order. The second field, called the admin status, contains the
status of the host specified in host identifier field. The format of
the admin status field is defined below.
Byte Bit Setting
0 0-7 'A' = accepted to receive
product
'D' = declined status confirmed
'P' = pending status
1 0-7 Decline reason
2 0-7 Unused, set to zero
3 0-7 Unused, set to zero
The usage of the Host List depends on Group type being used (also see
Group Management, Section 6). For Closed Groups, the Host List is
included in the message and contains the IP address of each Client
that is being invited to participate in the product transfer. The
Admin status is initially set to 'P'. As Clients send Registration
messages, the Server acknowledges receipt of the message by setting
the Admin status in the next Announce message that is transmitted.
When an invited Client indicates that it will be participating in the
product transfer, its Admin status is set to 'A' (accepted). When an
invited Client indicates that it is declining the product, its Admin
status is set to 'D' (declined) and the Decline Reason field is set to
the value received in the Registration message. If an uninvited Client
sends a Registration message for a Closed Group announcement, the
Server simply ignores the registration and sends an Abort to the
Miller, et. al. [Page 46]
INTERNET-DRAFT MFTP Specification January, 1997
client.
For Open Limited Groups, the Host List is initially omitted from the
Announce message. As Clients begin to send Registration messages, the
Server includes the Host List in subsequent Announce messages and adds
an entry for each Client that is accepted to participate in the
product delivery, and the Admin status is set to 'A'. If a host is
rejected by the Server (e.g. the maximum number of hosts that the
Server can handle is reached), its registration is simply ignored and
an entry is not placed in the Announce message. The Host List is never
included for Open Unlimited Groups.
12.2.2 Registration Message
The Registration message is sent only by the Client in order to
respond to a received Announce message. The Registration message is
sent to the Response Address contained in the Announce message. The
message contains the following parameters, which must appear in the
order shown.
Server Address (mandatory)
Server Port (mandatory)
Client Address (mandatory)
Registration Response (mandatory)
This response may be aggregated and relayed by an intermediate entity
to increase scalability. A single response that is not relayed
contains only the parameters shown. A relayed response repeats the
Client Address, Serial Number, and Registration Response parameters
for each Client that is being relayed.
Server Address
This field is echoed by the Client, exactly as received in the
Announce message. It may be used by an intermediate network entity
during the aggregation/relaying process.
Server Port
This field is echoed by the Client, exactly as received in the
Announce message. It may be used by an intermediate network entity
during the aggregation/relaying process.
Client Address
This field identifies the Client that is sending the response, and
identifies individual Clients in an aggregated/relayed response. The
address is in network byte order.
Registration Response
This parameter indicates the Client's intention to participate or not
participate in the transmission of a product announced by a Server. If
Miller, et. al. [Page 47]
INTERNET-DRAFT MFTP Specification January, 1997
the Client accepts the offer to participate in the transmission, the
reason for decline field has no meaning and is set to the value zero.
This parameter is a byte array with the following definition. Refer to
Section 13 for a list of the decline reasons.
Byte Bit Setting
0 0-7 'A' = accept
'D' = decline
1 0-7 Decline reason
2 0-7 Unused, set to zero
3 0-7 Unused, set to zero
12.2.3 Data Transfer Message
The Data Transfer message is transmitted only by the Server in order
to transfer the data product to the Clients. There is no response by
the Client to this message. This message may be sent in unicast,
broadcast, or multicast mode. In multicast mode, this message is sent
to the Private address. The message contains the following parameters.
Pass Number (mandatory)
Block Number (mandatory)
DTU Number (mandatory)
Product Data (mandatory)
Pass Number
This parameter contains the current pass number. The first pass is
numbered 1, and increments for each additional pass. This parameter
allows the Client to unambiguously determine the current pass number.
This is primarily useful for reporting reception status to an
application layer user.
Block Number
This parameter contains the sequence number of the current block,
within the current pass, starting at 1.
DTU Number
This parameter contains the sequence number of the current DTU, within
the current block, starting at 1.
Product Data
This parameter is the actual file data bytes being delivered to the
Client.
12.2.4 Status Request Message
The Status Request message is sent only by the Server in order to
query the Client for its reception status. The Client responds with an
Miller, et. al. [Page 48]
INTERNET-DRAFT MFTP Specification January, 1997
ACK or NAK Response message, depending on the type of status requested
(only NAKs are normally used). The Status Request message may be sent
in unicast, broadcast, or multicast mode. In multicast mode, it is
sent to the Private address. The message contains the following
parameters.
Pass Number (mandatory)
Block Range (mandatory)
DTU Range (mandatory)
Status Type (mandatory)
Host List (simple) (optional)
Pass Number
This parameter identifies the pass for which the Server is requesting
status. If there is no applicable pass number (e.g. the Server is
querying for status prior to a file restart), then this parameter
value is set to zero.
Block Range
This parameter identifies the range of blocks for which the Server is
requesting status. The parameter consists of two 32-bit fields. The
first field contains the number of the first block in the range, and
the second field contains the number of the last block in the range.
If status is being requested for a single block, both fields are set
to the same block number.
DTU Range
This parameter identifies the range of DTUs within a block for which
the Server is requesting status. The parameter consists of two 32-bit
fields. The first field contains the number of the first DTU in the
range, and the second field contains the number of the last DTU in the
range. The range cannot span multiple blocks. If status is being
requested for a single block, then this parameter may specify a subset
of all DTUs in the block. When status for multiple blocks is being
requested, this parameter is set to the first and last DTU numbers of
the entire block. That is, a DTU subset request is invalid when the
Server is requesting status for multiple blocks.
Status Type
This parameter is an integer that identifies the type of status
response that is being requested by the Server. The values are shown
below:
Miller, et. al. [Page 49]
INTERNET-DRAFT MFTP Specification January, 1997
1 = NAK responses only - The Client responds only if it recognizes
that it has not received one or more DTUs
in the block(s) specified by the Server.
It sends a NAK response message.
Otherwise, it sends no response at all.
This is the preferred mode of operation.
2 = ALL responses - The Client always responds regardless of
its reception status. If it has received
all DTUs in the indicated block, it sends
an ACK response message. Otherwise, it
sends a NAK response message.
Host List (simple)
Refer to the Join Group message for a description of this parameter.
It is included when the sender wants to specify Clients that should
respond to the message. If the message is to apply to all receiving
Clients, this parameter is omitted.
12.2.5 NAK Response Message
The NAK Response message is sent only by the Client when queried by
the Server via the Status Request message. The Client sends this
response if the Status Request message specifies either the ALL or NAK
response status type and the Client needs to request the
retransmission of one or more DTUs in the indicated block. The NAK
Response message is sent to the Response Address as specified in the
Announce message. The response message contains the following
parameters, which must appear in the order shown.
Server Address (mandatory)
Server Port (mandatory)
Client Address (mandatory)
Pass Number (mandatory)
Block Number (mandatory)
Event Count (mandatory)
Event Hash (mandatory)
Event List (mandatory)
Server Address
This field is echoed by the Client, exactly as received in the
Announce message.
Server Port
This field is echoed by the Client, exactly as received in the
Announce message.
Client Address
This field identifies the Client that is sending the response. The
address is in network byte order.
Miller, et. al. [Page 50]
INTERNET-DRAFT MFTP Specification January, 1997
Pass Number
This parameter is as specified for the Status Request message above.
That is, the received value is echoed in the response message.
Block Number
The value of this parameter is set to the block number that is being
reported. If the Server requests status for a single block, then this
parameter contains the same number. If the Server requests status for
a range of blocks, then a separate response message is returned for
each block in the range and this parameter is set to the block number
that is being reported in each message.
Event Count
This parameter indicates the total number of NAKed DTUs that are being
reported in this message. In other words, this is the number of bits
that are set in the Event List parameter.
Event Hash
This parameter is the result of a hash function being applied to the
Event List parameter. The purpose is for the receiver to be able to
easily determine if the Event List for any two or more Clients is
identical. The MD4 hash function (RFC 1320) is used. This is useful
for aggregation/relaying, when used.
Event List
This parameter identifies the DTUs that the Client has not received
for the block in the Block Number field. It is a bit map, where each
bit in the map represents a single DTU in the block. A bit set to '1'
means that the DTU has not been received and a bit set to '0' means
the DTU has been successfully received by the Client. Note that the
Server may request the status for a subset of the DTUs in a block via
the DTU Range parameter in the Status Request message. The Client
still sends a complete bit map, but sets the bits for all DTUs outside
the range to '0' (successfully received). Also note that the DTU Range
parameter is not echoed in the response message.
12.2.6 ACK Response Message
The ACK Response message is sent only by the Client when queried by
the Server via the Status Request message. The Client sends this
response if the Status Request message specifies the ALL response
status type and the Client has successfully received all of the DTUs
in the indicated block (Note that this is not the preferred mode of
operation). The ACK Response message is sent to the Response Address
as specified in the Announce message. The response message contains
the following parameters, which must appear in the order shown.
Miller, et. al. [Page 51]
INTERNET-DRAFT MFTP Specification January, 1997
Server Address (mandatory)
Server Port (mandatory)
Pass Number (mandatory)
Block Range (mandatory)
Client Address (mandatory)
Server Address
This field is echoed by the Client, exactly as received in the
Announce message.
Server Port
This field is echoed by the Client, exactly as received in the
Announce message.
Pass Number
This parameter is as specified for the Status Request message above,
i.e. the received value is echoed in the response message.
Block Range
This parameter identifies the range of blocks for which the Client is
reporting ACK status. The parameter consists of two 32-bit fields. The
first field contains the number of the first block in the range, and
the second field contains the number of the last block in the range.
If status is being reported for a single block, both fields are set to
the same block number.
Note that the Server may request the status of a range of blocks, and
the Client may return both ACK and NAK responses, according to the
status of each block. While NAK responses are sent individually for
each block being reported by the Client, the ACK response may report
the successful reception of a range of blocks.
Client Address
This field identifies the Client that is sending the response, and
identifies individual Clients in an aggregated response. The address
is in network byte order.
12.2.7 Done Message
The Done message is sent only by the Client to the Server to indicate
that a complete product has been received successfully. The message is
sent to the Response Address as specified in the Announce message. The
message contains the following parameters, which must appear in the
order shown.
Miller, et. al. [Page 52]
INTERNET-DRAFT MFTP Specification January, 1997
Server Address (mandatory)
Server Port (mandatory)
Client Address (mandatory)
Event Count (mandatory)
Data Transfer Duration (mandatory)
Server Address
This field is echoed by the Client, exactly as received in the
Announce message.
Server Port
This field is echoed by the Client, exactly as received in the
Announce message.
Client Address
This field identifies the Client that is sending the response. The
address is in network byte order.
Event Count
This parameter reports the total number of DTUs that the Client
reported as missed during reception of the product. In other words,
the field contains the total of all DTUs reported in NAK Response
messages sent to the Server during reception of the product.
Data Transfer Duration
This parameter reports the actual time, in seconds, that was required
for the Client to receive the product.
12.2.8 Completion Message
The Completion message is sent only by the Server in order to confirm
receipt of Done messages from Clients. When Clients are not sending
Done messages (i.e. Open-unlimited groups), the message announces the
end of a transmission. This message may be sent in unicast, broadcast,
or multicast mode. In multicast mode, this message is sent to the
Private address. The message contains the following parameter.
Host List (simple) (optional)
Host List
This parameter is included only when the Clients are sending Done
messages. The inclusion of a Client address in the list confirms
receipt of the Client's Done message.
Miller, et. al. [Page 53]
INTERNET-DRAFT MFTP Specification January, 1997
12.2.9 Abort Message
The Abort message is sent only by the Server in order to cancel a
product announcement or product delivery that is currently in progress
or to abort a specific Client or Clients. There is no response to
this message by the Client, so multiple Abort messages should be sent
to increase the probability of success. The message may be sent in
unicast, broadcast, or multicast mode. In multicast mode, the message
is sent to the Private address. The message contains the following
parameters.
Cancel Reason (mandatory)
Host List (simple) (optional)
Cancel Reason
This parameter indicates the reason that the Server is aborting a
product transmission. Refer to Section 13 for the value list.
Host List (simple)
Refer to the Join Group message for a description of this parameter.
This parameter is included when the Server wants to Abort a subset
of the Client population. If the parameter is omitted, then all
receiving Clients are affected.
12.2.10 Quit Message
The Quit message is sent only by the Client in order to terminate its
participation in the current product transfer. There is no response to
this message by the Server, so multiple messages should be sent to
increase the probability of success. The message is sent to the
Response Address as specified in the Announce message. It includes the
following parameters.
Cancel Reason (mandatory)
Cancel Reason
This parameter indicates the reason that a Client is terminating
reception of a product transmission. Refer to Section 11 for the
value list.
12.2.11 End Message
The End message is sent only by the Server in order to conveniently
indicate to aggregating entities in the network, if present, that the
session is over. End messages are sent four times after the Server
has determined the session is over on the Public Address. There are no
response messages back to the End message.
Miller, et. al. [Page 54]
INTERNET-DRAFT MFTP Specification January, 1997
The End message includes the following parameters.
Server Address (mandatory)
Server Port (mandatory)
Response Address (mandatory)
Response Port (mandatory)
13.Reason Codes
The various "reason" codes used in PDUs are listed below. Refer to
previous sections for a description of the message types and
configuration parameters mentioned here.
Client Registration Message - decline reasons
1001 insufficient disk space
1002 insufficient memory
1003 system error (disk, I/O, etc)
1004 already have file. The Server has indicated in the Announce
message that the Client should preserve any existing copy of
the file. Therefore, the Client is confirming that it already
has the file and will not be participating in the current
transfer. Refer to the overwrite/preserve file option in the
Announce message.
1005 no file to restart. The Server has indicated that this is a
file restart. However, the Client does not currently have any
portion of the file and the sender wants to restrict the
transfer to only those Clients that already have a portion of
the file. Refer to the initial/restart and late entry file
options in the Announce message.
1006 unknown application type (ARS). The Announce message specifies
an ARS that does not match the ARS set by the application
layer user. Refer to the Application Reference String
parameter in the Announce message.
1007 no interest in sender. The application layer user has not
specified this Server as one of the Servers that it wishes to
receive data products from.
1008 no interest in product type. The application layer user has
not specified this product type as one of the types that it
wishes to receive. Currently, only file product is supported
but this provides expansion capability for the future. Refer
to the product type specification in the Announce message.
Miller, et. al. [Page 55]
INTERNET-DRAFT MFTP Specification January, 1997
Client Quit Message - cancel reasons
2001 application action. The user application has terminated
this product transfer
2002 transfer timeout. The maximum time duration specified for the
transfer of this product has expired. Refer to the Data
Transfer Duration parameter in the Announce message.
2003 system error. An unrecoverable host error has occurred.
2004 error threshold. The error threshold has been reached for this
Client. Refer to the Error Threshold parameter in the Announce
message.
Server Abort Message - cancel reasons
3001 application action. The user application has terminated this
product transfer.
3002 transfer timeout. The maximum time duration specified for the
transfer of this product has expired. Refer to the Delivery
Time Limit configuration parameter.
3003 status timeout. The maximum time duration specified for status
acquisition has expired. Refer to the Status Retry Timer and
Status Retry Count configuration parameters.
3004 system error. An unrecoverable host error has occurred.
14. Future Extensions
Work is ongoing to provide improvements to the basic MFTP protocol.
Some of the improvements under consideration are:
1. More comprehensive flow control. As currently defined, MFTP
handles flow control simply by having client receivers with
excessive DTU error rates (excessive as defined by the server)
remove themselves from the group.
2. More generalized host lists. Currently, host lists include
the IP addresses of the hosts included in the list. It has
been suggested that the host lists contain the host names
rather than the IP address; however, these will occupy larger
fields than the 32 bit address field.
3. Addition of Security (see Section 15).
4. Interfaces to the mmusic protocols.
5. Interfacing to RSVP.
15. Security Considerations
This version of the protocol does not include any distinct message
types or additional fields for existing messages that will be needed
to support security. Security is currently under study and the
specification will be updated at a future time to address the security
requirements based on the work performed in the IPSEC IETF Working
Group.
Miller, et. al. [Page 56]
INTERNET-DRAFT MFTP Specification January, 1997
16. Acknowledgments
The authors would like to thank Scott Bradner (Harvard University),
Ken Cates (StarBurst Communications), and Tony Speakman (Cisco
Systems) for their comments and suggestions on various aspects of
this document.
References
[1] Deering, S., "Host Extensions for IP Multicasting", RFC 1112,
August 1989.
[2] Postel, J., Reynolds, J. "File Transfer Protocol (FTP)", RFC
959, October 1985.
[3] Schulzrinne, H., Casner, S., Frederick, R., Jacobson, V.,
"RTP: A Transport Protocol for Real-Time Applications", RFC
1889, January 1996.
[4] Schulzrinne, H.,"RTP Profile on Audio and Video Conferences
with Minimal Control", RFC 1890, January 1996.
Author's Addresses
Kenneth Miller Phone: +1 (508) 287-5560
StarBurst Communications, Inc. Email: miller@starburstcom.com
150 Baker Ave.
Concord, MA 01742
Kary Robertson Phone: +1 (508) 287-5560
StarBurst Communications, Inc. Email: krobertson@starburstcom.com
150 Baker Ave.
Concord, MA 01742
Alex Tweedly Phone: +1 (408) 526-8114
Cisco Systems, Inc. Email: agt@cisco.com
170 West Tasman Drive
San Jose, CA 95134
Marc White Phone: +1 (508) 287-5560
StarBurst Communications, Inc. Email: mwhite@starburstcom.com
150 Baker Ave.
Concord, MA 01742
Miller, et. al. [Page 57]