Internetworking Over NBMA                               James V. Luciani
INTERNET-DRAFT                                            (Bay Networks)
<draft-ietf-ion-scsp-mars-00.txt>                       Anthony M. Gallo
                                                                   (IBM)
                                                   Expires December 1997





                 A Distributed MARS Service Using SCSP


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 ``work in progress.''

   To learn the current status of any Internet-Draft, please check the
   ``1id-abstracts.txt'' listing contained in the Internet-Drafts Shadow
   Directories on ds.internic.net (US East Coast), nic.nordu.net
   (Europe), ftp.isi.edu (US West Coast), or munnari.oz.au (Pacific
   Rim).

Abstract

   This document describes a method for distributing a MARS service
   within a LIS[1].  This method uses the Server Cache Synchronization
   Protocol (SCSP)[2] to synchronize the MARS Server databases within a
   LIS.  When SCSP is used to synchronize the caches of MARS Servers in
   a LIS, the LIS defines the boundary of an SCSP Server Group (SG).

1. Introduction

   The MARS is an extended analog of the ATMARP Server introduced in
   [4]. It provides the necessary connection and addressing services
   required by layer 3 multicast services over ATM. There are three
   basic elements to the MARS model. First, the MARS Server which
   manages and distributes layer 3 group membership information to the
   LIS. Second, MARS Clients which register with and query a single MARS
   Server for layer 3 multicast information. Third, MCS Clients which
   register with a single MARS Server and provide layer 3 multicast
   forwarding services for a LIS.

   Both MARS Clients and MCS Clients explicitly register with the MARS
   Server before exchanging layer 3 multicast information. During the
   registration process MARS Clients are place on the Cluster Control VC
   (CCVC) and MCS Clients are placed on the Server Control VC (SCVC).
   Both the CCVC and SCVC are then used to propagate layer 3 multicast
   updates to the clients which make up a LIS. During the registration
   process MARS Clients are also assigned a unique Cluster Member ID
   (CMI) which is used to identify reflected packets in the presence of
   MCS Clients.

   In the Distributed MARS Model there MAY be multiple MARS Servers in a
   given LIS, and since any MARS Server within the LIS MUST be able to
   provide layer 3 multicast information about any multicast group
   within the LIS, there MUST be a method by which to synchronize
   multicast information across all MARS Servers within the LIS. In [5]
   several distributed MARS models are discussed along with various
   trade offs of each. The document provides a description of the
   problems that need to be addressed from the MARS protocol's point of
   view in a distributed system.

   The Server Cache Synchronization Protocol (SCSP) solves the
   generalized server synchronization/cache-replication problem for
   distributed databases, and thus SCSP MAY be applied to the MARS
   Server database synchronization problem within a LIS. When SCSP is
   used to synchronize the caches of MARS Servers in a LIS, the LIS
   defines the boundary of and SCSP Server Group (SG).

   SCSP is defined in two parts: the protocol independent part and the
   client/server protocol specific part. The protocol independent part
   is specified in [2] whereas this document will specify the
   client/server protocol specific part where the MARS Server is the
   client/server protocol.

2. Overview

   All MARS Servers belonging to a LIS are said to belong to a Server
   Group (SG).  A SG is identified by, not surprisingly, its SGID which
   is contained in a field in all SCSP packets. All SCSP packets contain
   a Protocol ID (PID) field as well. This PID field is set to 0x0003 to
   signify that SCSP is synchronizing MARS Server databases as opposed
   to synchronizing some other protocol's databases. (see Section
   B.2.0.1 of [2] for more details).  In general, PIDs for SCSP will be
   assigned by IANA upon request given that a client/server protocol
   specific specification has been written. In the case of MARS Servers,
   the client/server protocol specific specification was written at the
   same time time as SCSP, and thus a PID=0x0003 was assigned in [2].

   SCSP places no topological requirements upon a MARS Server SG.
   Obviously, however, the resultant graph of MARS Servers must span the
   set of MARS Servers being synchronized. For more information about
   the client/server protocol independent part of SCSP, the reader is
   encouraged to see [2].

   When a SG is using SCSP for synchronization, a MARS Client or MCS
   Client will register with only one MARS Server although it is allowed
   to choose any MARS Server in the SG for this registration). At
   registration time the MARS Client or MCS Client will be added to that
   MARS Servers respective CCVC or SCVC. Also, MARS Clients will be
   issued a unique CMI for the entire LIS. This document assumes at a
   minimum each MARS Server in the SG will be configured with a unique
   range of CMIs to assign to clients registering with that MARS Server.
   Use of some external means for allocating CMIs to MARS Servers in a
   SG is possible but beyond the scope of this document.

   When a MARS Client or MCS Client successfully registers with a MARS
   Server in the SG that MARS Server will propagate the registration
   information to its peer MARS Servers. The same propagation will occur
   for any subsequent group membership information learned from the
   clients. The peer MARS Server will then update its group membership
   database and propagate the information out its own CCVC or SCVC if
   needed.

   In the case of a MARS Server failure all peer MARS Servers in the SG
   MUST flush the client/group membership information learned from the
   failed MARS Server. The clients belonging to the failed MARS Servers
   CCVC and SCVC will migrate to the next available MARS Server as
   specified in Section 5.3 of [1]. When a client detects a failure of
   its MARS, it steps to the next backup MARS Server and attempts to
   register with the server. If the registration is successful the
   client will re-join all of its previous group membership information.
   If the registration fails, the process repeats until a functional
   MARS Server is found.

   Determining the operational state of a MARS Servers in a SG requires
   that each MARS Server send out an "alive" or "heartbeat" message
   similar to the MARS Redirect message sent out on the CCVC or SCVC for
   MARS Clients. However, this message will only be sent to MARS Servers
   in the SG and is from here on defined as the MARS Server Redirect
   Entry.

   In order to detect that a MARS Server failure has occurred each
   server MUST update it's MARS Server Redirect Entry state at least
   every 2 minutes, it is RECOMMENDED that it is updated every 1 minute.
   Failure to receive two consecutive MARS Server Redirect Entry updates
   from a given MARS Server in the SG will cause all membership
   information learned from this server to be flushed. The MARS Server
   Redirect Entry state is also used to create the MARS_REDIRECT_MAP
   messages sent out on CCVC for each MARS Server in the SG. The
   ordering of each server learned will be based on the MARS Servers
   SCSP Sender ID. The ordering of the MARS_REDIRECT_MAP will first
   contain the list of MARS Servers learned via MARS Server Redirect
   Entry updates in ascending order based on the SCSP Sender ID,
   followed by any externally configured or learned backup MARS Servers.

   In the case of a MARS Client or MCS Client failure where the client
   is unexpectedly removed from the CCVC or SCVC the MARS Server MUST
   notify its peer SG members via a proxy deregister for that client.
   Upon receiving a proxy deregister request from a peer SG member all
   membership information for the deregistering client MUST be removed.
   Any Clients sending multicast data to the failed client should also
   receive an unexpected removal of this client which will intern cause
   the sending client to revalidate the multicast groups current
   membership as outlined in Section 5.1.5.1 of [1].

3.  Format of the CSA Record MARS Specific Part

   CSA Records in SCSP contain a "Client/Server Protocol Specific Part"
   which contains the non-protocol independent information for a given
   server's cache entry.

   0                   1                   2                   3
   0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |        Hardware Type          |        Protocol Type          |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |                              SNAP                             |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     SNAP      |   Unused      |    Version    |   State       |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |            Flags              |       Cluster Member ID       |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   | Src Addr Len  | Group Addr Len| ATM Addr T/L  |ATM SubAddr T/L|
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     Source Protocol Address             (variable length)     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     Source ATM Address                  (variable length)     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     Source ATM SubAddress               (variable length)     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     Minimum Multicast Group Address     (variable length)     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
   |     Maximum Multicast Group Address     (variable length)     |
   +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

   Hardware Type
     Defines the type of "link layer" addresses being carried.  This
     value is the ATM Forum 'address family number' specified in [3] as
     15 decimal (0x000F).  This is the mar$afn field defined in [1].

   Protocol Type
     This field is the protocol type number for the protocol using MARS
     from [3].  (IPv4 is 0x0800).  This is the mar$pro.type field from
     [1].

   Protocol SNAP
     This field is the optional protocol SNAP extension to protocol
     type. This is the mar$pro.snap field from [1].

   Version Number
     0                    MARS Specific part of the CSA record.
     0x01                 Reserved for NHRP.
     0x02 - 0xEF          Reserved for future use by the IETF.
     0xF0 - 0xFE          Allocated for use by the ATM Forum.
     0xFF                 Experimental/Local use.
     Version Number for this document MUST be set to 0x00.

   State
     1   MARS Server Redirect Entry.
     2   MCS Register request.
     3   MARS Client Join/Register request.
     4   MARS Client Leave/Deregister request.
     5   MCS Deregister request.
     6   MCS Join request.
     7   MCS Leave request.

     All other State values should cause the CSA to be discarded.

   Src Addr Len
     This field contains the length of the Source Protocol Address
     field. For IPv4, the value is 4 if an address is specified. A null
     (non-existant) address MUST be coded as zero length, and no space
     allocated for it in the message body.  This is the mar$spln field
     from [1].

   Group Addr Len
     This field contains the length of the Group Protocol Address field.
     For IPv4, the value is 4 if an address is specified. A null (non-
     existant) address MUST be coded as zero length, and no space
     allocated for it in the message body.  This is the mar$tpln field
     from [1].

   ATM Addr T/L
     This field contains the type and length of the Source ATM Address
     field.  The type and length encodings are described in Section
     5.1.2 of [1].

   ATM SubAddr T/L
     This field contains the type and length of the Source ATM
     SubAddress field.  The type and length encodings are described in
     Section 5.1.2 of [1].

   Flags
     Th flags field is used to contain several flags and is similar to
     the mar$flags field from [1].
       mar$flags
          Bit 15   - mar$flags.layer3grp
          Bit 13   - mar$flags.register
          Bit 0-7  - mar$flags.sequence

       All remaining bits are reserved and MUST be zero. The
       mars$flags.sequence field is of local significance only to the
       Local Server (LS).

   Cluster Member CMI
     This field contains the CMI which uniquely identifies each endpoint
     within a LIS. This is the mar$cmi field from [1].

   Source ATM Address
     This is the Source's ATM address of an address binding in a MARS
     server cache entry.  The address, if specified, is E.164 or ATM
     Forum NSAPA. This is the mar$sha field from [1].

   Source ATM Subaddress
     This is the Source's ATM subaddress of an address binding in a MARS
     server cache entry.  The subaddress, if specified, is an ATM Forum
     NSAPA.  If null, no storage will be allocated. This is the mar$ssa
     field from [1].

   Minimum Multicast Group Address
     This is the internetwork address of the lower bound on the range of
     multicast group addresses for the address binding in a MARS server
     cache entry. This is the mar$min.N field from [1].

   Maximum Multicast Group Address
     This is the internetwork address of the upper bound on the range of
     multicast group addresses for the address binding in a MARS server
     cache entry. This is the mar$max.N field from [1].

4.  Values for SCSP Protocol Independent Part

   The following sections give values for fields of the SCSP Protocol
   Independent Part of the various SCSP messages.

4.1 Values for the SCSP "Mandatory Common Part"

   Protocol ID = 0x0003
   Sender ID Len = 0x04
   Recvr ID Len = 0x04

   See Section B.2.0.1 of [2] for a detailed description of these
   fields.

4.2 Values for the SCSP "CSAS Record"

   Cache Key Len = 0x04
   Orig ID Len = 0x04

   See Section B.2.0.2 of [2] for a detailed description of these
   fields.


5.  Detailed State Descriptions (Under Construction)


5.1  MARS Server Redirect Entry.

   TBD....

5.2  MCS Register request.

   TBD....

5.3  MARS Client Join/Register request.

   TBD....

5.4  MARS Client Leave/Deregister request.

   TBD....

5.5  MCS Deregister request.

   TBD....

5.6  MCS Join request.

   TBD....

5.7  MCS Leave request.

   TBD....


6.  Security Consideration

   Security is not addressed in this document but is addressed in the
   SCSP Protocol Independent part [2].


References

[1] "Support for Multicast over UNI 3.0/3.1 based ATM Networks", Armitage,
    RFC2022.

[2] "Server Cache Synchronization Protocol", Luciani, Armitage, Halpern,
    draft-ietf-ion-scsp-01.txt.

[3] Assigned Numbers, J. Reynolds and J. Postel, RFC 1700.

[4] "Classic IP and ARP over ATM", Mark Laubach, RFC 1577.

[5] "Redundant MARS architectures and SCSP, Armitage,
    draft-armitage-ion-mars-scsp-03.txt.


Acknowledgments

   The authors would like to thank Grenville Armitage for his previous
   distributed MARS work and also the members of the ION working group
   of the IETF, whose review and discussion of this document has been
   invaluable.


Author's Address

   James V. Luciani
   Bay Networks, Inc.
   3 Federal Street, BL3-04
   Billerica, MA  01821
   phone: +1-508-916-4734
   email: luciani@baynetworks.com

   Anthony M. Gallo
   IBM, Networking Hardware Division
   Dept. M6LA/B664
   P.O. Box 12195
   Research Triangle Park, NC 27709
   phone: +1-919-254-9889
   email: gallo@raliegh.ibm.com