INTERNET DRAFT              IS-IS restart                   Sept 2001

Network Working Group                                          M. Shand
Internet Draft                                            Cisco Systems
Expiration Date: February 2002
                                                              Sept 2001

                       Restart signaling for ISIS

Status of this Memo

   This document is an Internet-Draft and is in full conformance with
   all provisions of Section 10 of RFC2026 [1].

   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

   The list of Internet-Draft Shadow Directories can be accessed at

1. Abstract

   The IS-IS routing protocol (RFC 1142 [2], ISO/IEC 10589 [3]) is a
   link state intra-domain routing protocol. Normally, when an IS-IS
   router is re-started, the neighboring routers detect the restart
   event and cycle their adjacencies with the restarting router through
   the down state. This is necessary in order to invoke the protocol
   mechanisms to ensure correct re-synchronization of the LSP database.
   However, the cycling of the adjacency state causes the neighbors to
   regenerate their LSPs describing the adjacency concerned. This in
   turn causes temporary disruption of routes passing through the
   restarting router.

   In certain scenarios such temporary disruption of the routes is
   highly undesirable.

Shand                      Expires Feb 2002                   [Page 1]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   This draft describes a mechanism for a restarting router to signal
   that it is restarting to its neighbors, and allow them to re-
   establish their adjacencies without cycling through the down state,
   while still correctly initiating database synchronization.

   When such a router is restarted, it is highly desirable that it does
   not re-compute its own routes until it has achieved database
   synchronization with its neighbors. Re-computing its routes before
   synchronization is achieved will result in its own routes being
   temporarily incorrect.

   This draft additionally describes a mechanism for a restarting
   router to determine when it has achieved synchronization with its

2. Conventions used in this document

   The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
   this document are to be interpreted as described in RFC-2119 [4].

3. Overview

   There are two related problems with the existing specification of
   IS-IS with regard to re-synchronization of LSP databases when a
   router is re-started.

   Firstly, when a routing process restarts, and an adjacency to a
   neighboring router is re-initialized the neighboring routing process
   does three things

     1. It re-initializes the adjacency and causes its own LSP(s) to be
        regenerated, thus triggering SPF runs throughout the area (or
        in the case of Level 2, throughout the domain).

     2. It sets SRMflags on its own LSP database on the adjacency

     3. In the case of a Point-to-Point link it transmits a (set of)
        CSNP(s) over the adjacency.

   In the case of a restarting router process, the first of these is
   highly undesirable, but the second is essential in order to ensure
   re-synchronization of the LSP database.

   Secondly, whether or not the router is being re-started, it is
   desirable to be able to determine when the LSP databases of the
   neighboring routers have been synchronized (so that the overload bit
   can be cleared in the router's own LSP, for example). This document
   describes modifications to achieve this.

Shand                      Expires Feb 2002                   [Page 2]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   It is assumed that the three-way handshake [5] is being used on
   Point-to-Point circuits.

4. Approach

4.1 Timers

   A router that is restart capable maintains three additional timers,
   T1, T2 and T3.

   An instance of T1 is maintained per interface, and indicates the
   time after which an unacknowledged restart attempt will be repeated.
   A typical value might be 3 seconds.

   An instance of T2 is maintained for each LSP database present in the
   system. I.e. for a level1/2 system, there will be an instance of T2
   for Level 1 and one for level 2. This is the maximum time that the
   system will wait for LSPDB synchronization. A typical value might be
   60 seconds.

   A single instance of T3 is maintained for the entire system. It
   indicates the time after which the router will declare that it has
   failed to achieve database synchronization (by setting the overload
   bit in its own LSP). This is initialized to 65535 seconds, but is
   set to the minimum of the remaining times of received IIHs
   containing a restart TLV with RA set.

4.2 Adjacency re-acquisition

   Adjacency re-acquisition is the first step in re-initialization. The
   restarting router explicitly notifies its neighbor that the
   adjacency is being re-acquired, and hence that it should not re-
   initialize the adjacency. This is achieved by the inclusion of a new
   "re-start" option (TLV) in the IIH PDU. The presence of this TLV
   indicates that the sender supports the new restart capability and it
   carries flags that are used to convey information during a restart.
   All IIHs transmitted by a router that supports this capability MUST
   include this TLV.

     Type [TBD]
     Length 3
     Value (3 octets)
        Flags (1 octet)
               Bit 1 - Restart Request (RR)
               Bit 2 - Restart Acknowledgment (RA)
               Bits 3-8 รป Reserved
       Remaining Time (2 octets)
               Remaining holding time (in seconds)
               (note: only required when RA bit is set)

Shand                      Expires Feb 2002                   [Page 3]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   On receipt of an IIH with the "re-start" TLV having the RR bit set,
   if there exists on this interface an adjacency in state "Up" with
   the same System ID, and in the case of a LAN circuit, with the same
   source LAN address, then, irrespective of the other contents of the
   "Intermediate System Neighbors" option (LAN circuits), or the
   "Point-to-Point Adjacency State" option (Point-to-Point circuits):-

   a) DO NOT refresh the timer on the adjacency, but leave the
     adjacency in state "Up",

   b) immediately (i.e. without waiting for any currently running timer
     interval to expire, but with a small random delay of a few 10s of
     milliseconds on LANs to avoid "storms"), transmit over the
     corresponding interface an IIH including the "re-start" TLV with
     the RR bit clear and the RA bit set, having updated the "Point-to-
     Point Adjacency State" option to reflect any new values received
     from the re-starting router. (This allows the restarting router to
     quickly acquire the correct information to place in its hellos.)
     The "Remaining Time" MUST be set to the current time (in seconds)
     before the holding timer on this adjacency is due to expire,

   c) if the corresponding interface is a Point-to-Point interface, or
     if the receiving router has the highest LnRouterPriority (with
     highest source MAC address breaking ties) among those routers
     whose IIHs contain the restart TLV, excluding the transmitting
     router (note the actual DR is NOT changed by this process.),
     initiate the transmission over the corresponding interface of a
     complete set of CSNPs, and set SRMflags on the corresponding
     interface for all LSPs in the local LSP database.

   Otherwise (i.e. if there was no adjacency to the system ID in
   question), process the IIH as normal by re-initializing the
   adjacency, and setting the RA bit in the returned IIH.

   A router that does not support the re-start capability will ignore
   the "re-start" TLV and re-initialize the adjacency as normal,
   returning an IIH without the "re-start" TLV.

   On starting, a router initializes the timer T3, starts timer T2 for
   each LSPDB and for each interface starts a timer T1 and transmits an
   IIH containing the "re-start" TLV with the RR bit set.

      1. On a LAN circuit the IIH contains an empty "Intermediate
         Systems Neighbors" TLV.

      2. On a Point-to-Point circuit the IIH contains a "Point-to-Point
         Adjacency State" option with state "Down", and with empty
         "Neighbor System ID" and "Neighbor Extended Local Circuit ID"
         options. The values of the "LocalCircuitID" and the "Extended
         Local CircuitID" may, but need not be, the same as those used
         previously for this circuit.

Shand                      Expires Feb 2002                   [Page 4]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   Transmission of "normal" IIHs is inhibited until the conditions
   described below are met (in order to avoid causing an unnecessary
   adjacency re-initialization). On expiry of the timer T1, it is
   restarted and the IIH is re-transmitted as above.

   On receipt of an IIH by the restarting router, a local adjacency is
   established as usual, and if the IIH contains a "re-start" TLV with
   the RA bit set, the receipt of the acknowledgement over that
   interface is noted (see section Error! Reference source not found.).
   T3 is set to the minimum of its current value and the value of the
   "Remaining Time" field in the received IIH.

   Receipt of an IIH not containing the "re-start" option is also
   treated as an acknowledgement, since it indicates that the neighbor
   is not re-start capable. In this case the neighbor will have re-
   initialized the adjacency as normal, which in the case of a Point-
   to-Point link will guarantee that SRMflags have been set on its
   database, thus ensuring eventual LSPDB synchronization. In the case
   of a LAN interface, the usual operation of the update process will
   also ensure that synchronization is eventually achieved. However,
   since no CSNP is guaranteed to be received over this interface, T1
   is cancelled immediately without waiting for a CSNP. Synchronization
   may therefore be deemed complete even though there are some LSPs
   which are held (only) by this neighbor (see section 4.3).

   In the case of a Point-to-Point circuit, the "LocalCircuitID" and
   "Extended Local Circuit ID" information contained in the IIH can be
   used immediately to generate an IIH containing the correct 3-way
   handshake information. The presence of "Neighbor System ID" or
   "Neighbor Extended Local Circuit ID" information which does not
   match the values currently in use by the local system is ignored
   (since the IIH may have been transmitted before the neighbor had
   received the new values from the re-starting router), but the
   adjacency remains in the initializing state until the correct
   information is received.

   In the case of a LAN circuit the information in the Intermediate
   Systems Neighbors option is recorded and used for the generation of
   subsequent IIHs as normal.

   When BOTH a complete set of CSNP(s) and an acknowledgement have been
   received over the interface, the timer T1 is cancelled.

   Once T3 has expired or been cancelled, subsequent IIHs are
   transmitted according to the normal algorithms, but including the
   "re-start" TLV with both RR and RA clear.

   If a LAN contains a mixture of systems, only some of which support
   the new algorithm, database synchronization is still guaranteed, but
   the "old" systems will have re-initialized their adjacencies.

Shand                      Expires Feb 2002                   [Page 5]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   If an interface is active, but does not have any neighboring router
   reachable over that interface the timer T1 would never be cancelled,
   and according to clause the SPF would never be run.
   Therefore timer T1 is cancelled after some pre-determined number of
   expirations. (By this time any existing adjacency on a remote system
   would probably have expired anyway.)

   A router which supports re-start SHOULD ensure that the holding time
   of any IIHs it transmits is greater than the expected time to
   complete a re-start.

4.3 Database synchronization

   When a router is started or re-started it can expect to receive a
   (set of) CSNP(s) over each interface. The arrival of the CSNP(s) is
   now guaranteed, since the "re-start" IIH with the RR bit set will be
   retransmitted until the CSNP(s) are correctly received.

   The CSNPs describe the set of LSPs that are currently held by each
   neighbor. Synchronization will be complete when all these LSPs have
   been received.

   On starting, a router starts the timer T3 and an instance of timer
   T2 for each LSPDB. In addition to normal processing of the CSNPs,
   the set of LSPIDs contained in the first complete set of CSNP(s)
   received over each interface is recorded, together with their
   remaining lifetime. If there are multiple interfaces on the
   restarting router, the recorded set of LSPIDs is the union of those
   received over each interface. LSPs with a remaining lifetime of zero
   are NOT so recorded.

   As LSPs are received (by the normal operation of the update process)
   over any interface, the corresponding LSPID entry is removed (it is
   also removed if the LSP had arrived before the CSNP containing the
   reference). When an LSPID has been held in the list for its
   indicated remaining lifetime, it is removed from the list. When the
   list of LSPIDs becomes empty, the timer T2 is cancelled.

   At this point the local database is guaranteed to contain all the
   LSP(s) (either the same sequence number, or a more recent sequence
   number) which were present in the neighbors' databases at the time
   of re-starting. LSPs that arrived in a neighbor's database after the
   time of re-starting may, or may not, be present, but the normal
   operation of the update process will guarantee that they will
   eventually be received. At this point the local database is deemed
   to be "synchronized".

   Since LSPs mentioned in the CSNP(s) with a zero remaining lifetime
   are not recorded, and those with a short remaining lifetime are
   deleted from the list when the lifetime expires, cancellation of the
   timer T2 will not be prevented by waiting for an LSP that will never

Shand                      Expires Feb 2002                   [Page 6]

INTERNET DRAFT              IS-IS restart                   Sept 2001

4.3.1 LSP generation and flooding and SPF computation

   The operation of a router starting, as opposed to re-starting is
   somewhat different. These two cases are dealt with separately below. Starting for the first time

   In the case of a starting router, as soon as each adjacency is
   established, and before any CSNP exchanges, the router's own zeroth
   LSP is transmitted with the overload bit set. This prevents other
   routers from computing routes through the router until it has
   reliably acquired the complete set of LSPs. The overload bit remains
   set in subsequent transmissions of the zeroth LSP (such as will
   occur if a previous copy of the routers LSP is still present in the
   network) while any timer T2 is running.

   When all the T2 timers have been cancelled, the own LSP(s) are
   regenerated with the overload bit clear (assuming the router isn't
   in fact overloaded), and flooded as normal.

   Other 'own' LSPs (including pseudonodes) are generated and flooded
   as normal, irrespective of the timer T2. The SPF is also run as
   normal and the RIB and FIB updated as routes become available. Re-starting

   In order to avoid causing unnecessary routing churn in other
   routers, it is highly desirable that the own LSPs generated by the
   restarting system are the same as those previously present in the
   network (assuming no other changes have taken place). It is
   important therefore not to regenerate and flood the LSPs until all
   the adjacencies have been re-established and any information
   required for propagation into the local LSPs is fully available.
   Ideally, the information should be loaded into the LSPs in a
   deterministic way, such that the same information occurs in the same
   place in the same LSP (and hence the LSPs are identical to their
   previous versions). If this can be achieved, the new versions will
   not even cause SPF to be run in other systems. However, provided the
   same information is included in the set of LSPs (albeit in a
   different order, and possibly different LSPs), the result of running
   the SPF will be the same and will not cause churn to the forwarding

   In the case of a re-starting router, none of the router's own non-
   pseudonode LSPs are transmitted, nor is the SPF run to update the
   forwarding tables while the timer T3 is running.

   Redistribution of inter-level information must be regenerated before
   this router's LSP is flooded to other nodes. Therefore the level-n
   non-pseudonode LSP(s) should not be flooded until the other level's
   T2 timer has expired and its SPF has been run. This ensures that any

Shand                      Expires Feb 2002                   [Page 7]

INTERNET DRAFT              IS-IS restart                   Sept 2001

   inter-level information that should be propagated can be included in
   the level-n LSP(s).

   During this period, if one of the router's own (including
   pseudonodes) LSPs is received, which the local router does not
   currently have in its own database, it is NOT purged. Under normal
   operation, such an LSP would be purged, since the LSP clearly should
   not be present in the global LSP database. However, in the present
   circumstances, this would be highly undesirable, because it could
   cause premature removal of an own LSP -- and hence churn in remote
   routers. Even if the local system has one or more own LSPs (which it
   has generated, but not yet transmitted) it is still not valid to
   compare the received LSP against this set, since it may be that as a
   result of propagation between level 1 and level 2 (or vice versa) a
   further own LSP will need to be generated when the LSP databases
   have synchronized.

   When the timer T2 expires, or is cancelled, the SPF is run to update
   the RIB and FIB.

   Once the other level's SPF has run and any inter-level propagation
   has been resolved, the 'own' LSPs can be generated and flooded. Any
   'own' LSPs which were previously ignored, but which are not part of
   the current set of 'own' LSPs (including pseudonodes) should then be
   purged. Note that it is possible that a Designated Router change may
   have taken place, and consequently the router should purge those
   pseudonode LSPs which it previously owned, but which are now no
   longer part of its set of pseudonode LSPs.

   If the timer T3 expires before all the T2 timers have expired, this
   indicates that the synchronization process is taking longer than
   minimum holding time of the neighbors. The router's own LSP(s) for
   levels which have not yet completed their first SPF computation are
   then flooded with the overload bit set to indicate that the router's
   LSPDB is not yet synchronized (and other routers should therefore
   not compute routes through this router). In order to prevent the
   neighbor's adjacencies from expiring, IIHs are transmitted over all
   interfaces with neither RR nor RA set in the restart TLV. This will
   cause the neighbors to refresh their adjacencies. The own LSP(s)
   will continue to have the overload bit set until timer T2 has been
   cancelled as in the case of starting for the first time described in

5. Security Considerations

   This memo does not create any new security issues for the IS-IS
   protocol. Security considerations for the base IS-IS protocol are
   covered in [2] and [3].

Shand                      Expires Feb 2002                   [Page 8]

INTERNET DRAFT              IS-IS restart                   Sept 2001

6. References

   1  Bradner, S., "The Internet Standards Process -- Revision 3", BCP
      9, RFC 2026, October 1996.

   2  Callon, R., "OSI IS-IS for IP and Dual Environment," RFC 1195,
      December 1990.

   3  ISO, "Intermediate system to Intermediate system routeing
      information exchange protocol for use in conjunction with the
      Protocol for providing the Connectionless-mode Network Service
      (ISO 8473)," ISO/IEC 10589:1992.

   4  Bradner, S., "Key words for use in RFCs to Indicate Requirement
      Levels", BCP 14, RFC 2119, March 1997

   5  Katz, D., "Three-Way Handshake for IS-IS Point-to-Point
      Adjacencies", draft-ietf-isis-3way-03.txt, July 2000

7. Acknowledgments

   The author would like to acknowledge contributions made by Radia
   Perlman, Mark Schaefer, Russ White and Rena Yang.

8. Author's Addresses

   Mike Shand
   Cisco Systems
   4, The Square,
   Stockley Park,
   UB11 1BN, UK

   Phone: +44 20 8756 8690

Shand                      Expires Feb 2002                   [Page 9]