INTERNET DRAFT V. Kashyap
<draft-kashyap-ipoib-connected-mode-00.txt> IBM
Expiration Date: August 2003 February 2003
IP over InfiniBand: Connected Mode
Status of this memo
This document is an Internet-Draft and is in full conformance
with all provisions of Section 10 of RFC 2026.
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 ``work in progress''.
The list of current Internet-Drafts can be accessed at
http://www.ietf.org/ietf/1id-abstracts.txt
The list of Internet-Draft Shadow Directories can be accessed
at http://www.ietf.org/shadow.html
This memo provides information for the Internet community.
This memo does not specify an Internet standard of any kind.
Distribution of this memo is unlimited.
Copyright Notice
Copyright (C) The Internet Society (2001). All Rights Reserved.
Abstract
The InfiniBand Architecture(IBA) defines a high speed, channel
based interconnect between systems and devices. IBA provides
multiple modes of transport services with differing
characteristics. This document describes IP over IBA's Connected
transport modes.
Kashyap [Page 1]
INTERNET-DRAFT Connected mode IPoIB February 2003
Table of Contents
1.0 Introduction
2.0 IPoIB-connected mode
2.1 Outline of Address Resolution
2.2 Outline of Connection Setup
3.0 Address Resolution
4.0 Connection setup
4.1 Service ID
4.2 MTU
5.0 IP Encapsulation
6.0 Security Considerations
7.0 References
1.0 Introduction
IBA defines two connected modes:
1. Reliable Connected(RC)
2. Unreliable Connected(UC)
The two modes differ mainly, as is clear from the names, in
providing reliability of data delivery across the connection.
However, both these modes will be considered together in this
document since all the discussion applies equally to both the
modes - the two modes are referred to as IPoIB-CM (connected
mode) in this document. IPoIB over reliable connected mode is
referred to as IPoIB-RC whereas IPoIB over unreliable connected
mode is referred to as IPoIB-UC.
The connected modes offer link MTUs of upto 2^31 bytes in length.
Thus the use of connected modes can offer significant benefits by
supporting reasonably large MTUs. The datagram modes are limited
to 4096 bytes. Reliability is also enhanced if the underlying
feature of 'automatic path migration' supported by the connected
modes is utilised [IBARCH].
This document presents a method of address resolution and
transmission of IP packets over connected modes of IBA.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL
NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL"
in this document are to be interpreted as described in RFC 2119.
Kashyap [Page 2]
INTERNET-DRAFT Connected mode IPoIB February 2003
2.0 IPoIB-connected mode
The connected modes of IBA do not support multicasting though
every node is capable of communicating i.e. setting a connection,
to every other node in the InfiniBand(IB) network.
This implies that intrinsically one cannot rely on native
broadcast or multicast to send out the address resolution query.
But an ARP server is not an efficient solution. Fortunately in the
case of IPoIB-CM there is a way out.
2.1 Outline of Address Resolution
IBA requires that all Host Channel Adapters(HCAs) support the
reliable and unreliable connected modes. It is optional for Target
Channel Adapters (TCAs) to supported the connected modes. At the
same time IBA requires all HCAs and TCAs to support unreliable
datagram mode. The unreliable datagram mode does support
multicasting. IPoIB over unreliable datagram(IPoIB-UD) as
presented in [IPoIB_MCAST]/[IPoIB_ENCAP] requires the IB subnet to
support IB level multicast.
The above makes it possible to use a multicast query for IPoIB-CM
address resolution. The address resolution in IPoIB-CM networks
uses an unreliable datagram(UD) queue pair(QP).
IPoIB_ENCAP proposes that the address resolution query is
multicast over an IB multicast address that is joined by every
member of the IPoIB subnet. This IB multicast address is referred
to as the 'broadcast-GID' [IPoIB_ENCAP]. This document extends the
requirement of joining the 'broadcast-GID' to IPoIB-CM too by
associating an unreliable datagram with it.
A broadcast-GID is formed with the knowledge of the scope bits, IP
version and the partition key(P_Key) associated with the subnet.
Thus these three parameters must be known to the node before an
IPoIB interface can be brought up. The exact format and rules to
setup the broadcast-GID are defined in [IPoIB_MCAST].
An implementation MAY use the same unreliable datagram(UD) queue
pair(QP) as used by the IPoIB-UD implementation if the latter mode
is supported in the same partition and scope.
Therefore during address resolution, the query is sent out on the
broadcast GID. A unicast reply is received on the UD QP associated
with IPoIB-CM.
Kashyap [Page 3]
INTERNET-DRAFT Connected mode IPoIB February 2003
2.2 Outline of Connection setup
Address resolution is but the first step. Once the link address of
the remote node is known a connection must be setup between the
nodes before any IP communication may occur.
To make a connection, the sender must know the service-ID to use
in the request to make a connection [IBARCH]. It must also supply
the queue pair to the remote node. The peer replies with its queue
pair. Note that every connection is peer to peer and does not use
shared queue pairs(QPs). Every connection uses a pair of unique
(on the node) QPs.
Though the address resolution occurs at an individual address
level the connection between the nodes is at the IB layer and so,
every individual address resolution does not imply a new connection
between the peers.
3.0 Address Resolution
Every IPoIB-CM node MUST join the broadcast-GID associated with the
subnet. The address resolution query is always sent out on the
broadcast-GID.
IPoIB encapsulation [IPoIB_ENCAP] describes the link-layer address
as follows:
<1 octet reserved>:QP: GID
This document extends the link-layer address as follows:
<flags>:QP-cookie:GID
QP-cookie: An IB connection, as noted above, uses a pair of QPs;
one on each node. An implementation therefore cannot
advertise the QP it intends to use as part of the
address resolution query since such a message is seen
by all members of the subnet. Instead an
implementation specific, 3-octet long, QP-cookie is
used.
Kashyap [Page 4]
INTERNET-DRAFT Connected mode IPoIB February 2003
The QP-cookie is used by the peer for two purposes:
a. The QP-cookie is used to form the service-ID that
is used in the IB connection messages. The
receiver can, depending on the service-ID and
thereby the QP-cookie make a decision on the QPN
to create (or deny).
b. The QP-cookie is used to determine if a connection
already exists to the peer. In such a case the
node can avoid an attempt at connection setup.
Note that the GID cannot be used for this purpose
since a GID can always be shared by multiple
interfaces.
Flags: This is a single octet field. If bit 0 is set then it
implies that in the sender's view, the subnet is built
over IB's 'reliable connected' i.e. RC mode. If bit 1
is set then it implies that the subnet is built over
IB's 'unreliable connected' i.e. UC mode. All other
bits in the octet are reserved and MUST be set to 0.
Both the RC and UC flags MUST not be set at the same
time. They are mutually exclusive.
The receiver replies with its own link-address and the set of
flags. If the flags do not match then there is a misconfiguration
since the members of the same expect different link
characteristics (IB modes). In such a case a suitable error
indication SHOULD be provided to the administrator.
The receiver's reply is unicast back to the sender after the
receiver has, as in the case of IPoIB over unreliable
datagram(IPoIB_UD), after it resolves the GID to the LID.
Once the address resolution is completed the connection may be
setup.
4.0 Connection Setup
The connection may be setup by any of the peers though it is more
likely that the one that initiated the address resolution phase,
probably as a result of the need to send IP data, will initiate
the connection setup. IBA allows passive-active and active-active
Kashyap [Page 5]
INTERNET-DRAFT Connected mode IPoIB February 2003
connection setup.
The service ID used when setting up the connection is derived from
the QP cookie received during the address resolution process. A
node MAY return the same cookie for multiple addresses. For
example, the node might support multiple subnets over the same GID
and prefer to make only one IB level connection. This includes
supporting the same IB connection for both IPv4 and IPv6 to a peer
node. The choice of the QP-cookie is implementation dependent.
Therefore, the end initiating the connection needs to defer to the
peer's choice. If the peer has returned the same QP-cookie as
a result of multiple address resolution requests then, for those
addresses, a connection attempt SHOULD not be made. This is true
even if the requestor itself presented different QP-cookies, and
would have created separate QPs if it had received the request
instead.
If a node does receive a connection request for the same
service-ID from the same peer then it is upto the implementation
to honour or reject it.
4.1 Service ID
The InfiniBand specification defines a block of service IDs
for IETF use. The InfiniBand specification has left the
definition and management of this block to the IETF. The
64-bit block is:
+--------+--------+--------+--------+--------+--------+--------+--------+
|00000001|<-------------------IETF use--------------------------------->|
+--------+--------+--------+--------+--------+--------+--------+--------+
The ServiceIDs used by IPoIB will use the following format:
+--------+--------+--------+--------+--------+--------+--------+--------+
|00000001|Reserved| 3-octet QP cookie |Reserved|Reserved|Reserved|
+--------+--------+--------+--------+--------+--------+--------+--------+
The Reserved fields MUST be transmitted as zeroes. They are
ignored on reception.
4.2 MTU
The IB connection might be used for both IPv4 and IPv6 or it
could be used for only one of them while a different connection is
used for the other. If the connection is used for both IPv4 and
IPv6, the link MTU MUST be able to support the minimum MTU
Kashyap [Page 6]
INTERNET-DRAFT Connected mode IPoIB February 2003
required by both IPv4 and IPv6.
Every connection setup message includes a 'private data' field
[IBARCH]. The private data field MUST carry the following
information:
0 15
+----------------+
| Desired MTU |
+----------------+
| Minimum MTU |
+----------------+
The connection setup message (CM REQ) MUST insert the requested
MTU in the 'Desired MTU' field and the minimum acceptable MTU. If
it is not acceptable to the peer then it MUST indicate the
preferred value in the 'desired MTU' when rejecting (CM REJ) the
request. If the 'desired MTU' is lower than the minimum MTU that
can be supported, the connection MUST be rejected (CM REJ message)
with the minimum acceptable MTU in both the desired and minimum
MTU fields.
5.0 IP encapsulation
The IP encapsulation will be done as defined in the IPoIB
encapsulation standard[IPoIB_ENCAP].
IP multicast cannot be done over the IPoIB-CM modes.
6.0 Security Considerations
A node may be returned a false set of flags by an imposter. This
may cause unnecessary attempts and some delay/disruption in IPoIB
communication. The same is the case if wrong/spurious QP-cookie
values are provided.
Kashyap [Page 7]
INTERNET-DRAFT Connected mode IPoIB February 2003
7.0 References
[IB_ARCH] InfiniBand Architecture Specification, version 1.1
www.infinibandta.org
[IPoIB_ARCH] draft-ietf-ipoib-architecture-01.txt, V. Kashyap
[IPoIB_ENCAP] draft-ietf-ipoib-ip-over-infiniband-01.txt,
V. Kashyap, H.K. Jerry Chu
[IPoIB_MCAST] draft-ietf-ipoib-link-multicast-02.txt,
H.K. Jerry Chu, V. Kashyap
7.0 Author's Address
Vivek Kashyap
15450, SW Koll Parkway
Beaverton, OR 97006
Phone: +1 503 578 3422
Email: vivk@us.ibm.com
Full Copyright Statement
Copyright (C) The Internet Society (2001). All Rights Reserved.
This document and translations of it may be copied and
furnished to others, and derivative works that comment on or
otherwise explain it or assist in its implementation may be
prepared, copied, published and distributed, in whole or in
part, without restriction of any kind, provided that the above
copyright notice and this paragraph are included on all such
copies and derivative works. However, this document itself may
not be modified in any way, such as by removing the copyright
notice or references to the Internet Society or other Internet
organizations, except as needed for the purpose of developing
Internet standards in which case the procedures for copyrights
defined in the Internet Standards process must be followed, or
as required to translate it into languages other than
English.
The limited permissions granted above are perpetual and will
not be revoked by the Internet Society or its successors or
assigns.
This document and the information contained herein is provided
on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET
Kashyap [Page 8]
INTERNET-DRAFT Connected mode IPoIB February 2003
ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE
USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR
ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A
PARTICULAR PURPOSE.
Kashyap [Page 9]