You can configure IS-IS to react more rapidly to isolated events that are likely to be real link failures and to react more
stably to frequent events that are unlikely to be actual link failures. The convergence speed and stability of IS-IS is affected
by the values that you set for various throttling timers. The throttling timers impose a trade-off between reaction time to
external events and the amount of resources dedicated to maintaining the information in the Routing Information Base (RIB).
You should become familiar with the following.
IS-IS PDUs
IS-IS encapsulates data into a data-link protocol data unit (PDU). There are four different PDU types and each can be Level
1 or Level 2:
-
LSP
--An LSP is a PDU that is sent between two IS-IS neighbors. The LSP contains information about neighbors and path costs, including
adjacencies to neighbors, connected IP prefixes, Open Systems Interconnection (OSI) end systems, and area addresses. LSPs
are used by the receiving routers to maintain their routing tables.
-
IIH
--An IS-IS Hello PDU is used to establish and maintain adjacencies. By default, an Intermediate-to-Intermediate Hello (IIH)
is padded to the maximum transmission unit (MTU) size.
-
PSNP
--A partial sequence number PDU (PSNP) contains summaries of only a subset of known LSPs. A PSNP is used to acknowledge and
request link-state information by soliciting newer versions of a complete LSP, or acknowledging receipt of an updated LSP,
respectively.
-
CSNP
--A complete sequence number PDU (CSNP) contains summaries of all LSPs known by the issuing router.
LSP-Related Intervals and Exponential Backoff Timers
The following timers and intervals relate to LSPs that are generated by the IS-IS router.
-
LSP
refresh
interval
--Specifies the number of seconds (0 to 65535) the router will wait before refreshing (re-creating and reflooding) its own
LSP.
-
Maximum
LSP
lifetime
--Specifies the value of the lifetime in the LSP header. Lifetime is used by all IS-IS routers in order to age out and purge
old LSPs.
The following exponential backoff timers have been implemented in IS-IS to control the events of SPF calculation, Partial
Route Calculations (PRC) computation, and LSP generation:
-
PRC
interval
--Specifies the number of seconds between two consecutive PRCs. When changes that do not affect the topology, such as advertised
external prefixes, are detected, the PRC is triggered.
-
LSP
generation
interval
--Specifies the number of seconds between creating new versions of a given LSP on a per-node basis.
-
SPF
interval
--Specifies the number of seconds between two consecutive SPF calculations.
The purpose of these exponential backoff timers is to react quickly to the first events but, under constant churn, to slow
down in order to prevent the CPU of the router from collapsing. The exponential backoff algorithm operates as follows:
-
An initial event triggers the SPF, PRC, or LSP generation.
-
The initial wait time that is configured for the interval determines the time between the initial event and the start of the
SPF, PRC, or LSP generation.
-
The incremental wait time that is configured for the interval determines the amount of time that the router will wait in between
the consecutive SPF execution, PRC execution, or LSP generation. This incremental value will increase exponentially between
the incremental events until the maximum value is reached. For example, the incremental value will be (1x incremental value)
between the first and second events, (2 x incremental value) between the second and third event, (4 x incremental value) between
the third and fourth event, (8 x incremental value) between the fourth and fifth event, and so on, until the configured maximum
interval--amount of time in seconds that the router will wait in between consecutive SPF execution, PRC execution, or LSP
generation--has been reached.
-
If no new triggers have been received after two times the configured maximum wait-interval value, the network stabilizes,
returning to a steady state and fast behavior. The initial wait-time interval will be reinstated.
See the to configure the recommended settings for the SPF, PRC and LSP generation timers.
IS-IS Hello PDU Timers
The different IS-IS Hello timers need to be adapted according to the adjacency convergence time required for each subnet.
Where a rapid adjacency loss has been detected, the timers need to be reduced. These timers should be modified if necessary
after deployment and after an accurate monitoring of the network stability and convergence has occurred.
-
Hello
interval
--Number of seconds during two consecutive transmissions of IIH PDUs.
-
Hello
interval
minimum
--When the hello interval is configured, the hold time is set to one second. The significance of the hello multiplier changes
if Fast Hellos are used; the hello multiplier becomes the number of hellos that will be sent per second.
-
Hello
multiplier
--An integer from 1 to 300 that is used to calculate the hold time. The hold time is the number of seconds during which the
router will wait for an IIH before declaring that its neighbor is lost. The router multiplies the hello interval by the hello
multiplier to determine the hold time. To avoid unnecessary adjacency resets, increase the default value of 3 on interfaces
where frequent losses of IIH PDUs are detected.
-
IS-IS
retransmit
interval
--Specifies the number of seconds between the resending of IS-IS link-state PDU transmissions for point-to-point links.
CSNP Interval
TheCSNP interval specifies the number of seconds between the two consecutive transmissions of CSNP PDUs. CSNP are generated
by the designated router (DIS) in order for all routers connected to a broadcast media to synchronize their databases and
by adjacent routers on a point-to-point network while setting up an adjacency. CSNPs are used to keep all router databases
up to date. The lower the value of the CSNP interval, the faster the speed of the sychronization. However, a CSNP interval
that is too low will trigger intensive PSNP PDU transmissions. All routers that are not synchronized with the DIS (Designated
Intermediate System) and that, therefore, need additional LSPs in their database send PSNPs.
SPF, PRC, and LSP generation exponential backoff timers need to be tuned according to the level of stability of the network
and the stability required in the routing domain. For instance, setting low values will trigger a fast convergence with a
potential risk of high resource utilization if flapping routes cause network churn. Setting high values will keep the network
stable with slower convergence.
It is recommended to leave the default value for the LSP generation interval at 5 seconds and also to increase the maximum
lifetime for LSPs to 65,535 seconds, in order to conserve CPU usage for generation and refreshing of LSPs.
If you are using a routing algorithm based on SPF and if you use values for the initial required delay that are fewer than
40 milliseconds, SPF may start before the LSP that triggered SPF is flooded to neighbors. The router should always flood,
at least, the LSP that triggered SPF before the router runs the SPF computation. LSP flooding is required in order to guarantee
that the network update in the LSP is propagated around the network as quickly as possible.