GSR Route Reflector: maximum number of clients?

Mailman List Mirror (Read Only)
1

I have a GSR 12008 that is acting as a route reflector and core router.
What is the practical maximum (from an operational standpoint) number of
clients it can support? (My Cisco sources tell me that the maximum
technically possible is large and does not significantly impact the
operation the forwarding of the GSR.)

In terms of manageability and simplicity, is it more favourable to have a
separate box (e.g. 7206 VXR) handle the route reflector job?

Thanks in advance for your thoughts.

James

2

Something like 40 iBGP peering sessions should be doable in most
  circumstances. Mileage will vary significantly as a function
  of:

    physical topology
    RR Topology
    policy (routes, dampening, etc..)
    stability
    .. and other things

  Adding a tertiary box such as a 7206 or some such box as a BGP
  route reflector may work fine -- but can give rise to improper
  routing information. (esp. wrt route selection)

  In general, one wants the routing information (control plane)
  to follow the forwarding path (data plane) as closely as
  possible.

  -alan

Thus spake James DeMong (James.DeMong@telus.com)
on or about Tue, May 01, 2001 at 01:31:27PM -0600:

3

Just to add to Alan's 'stability' bullet, one other thing to bear in mind
would be convergence time of routes on any given box in a reflected
'region' during an event if said multi-use route-reflector is attempting
to do other useful things, such as shovel packets, at the same time as
converge routes with all it's clients.

IME, you will need to decide on the importance to you of reconverging
during/following a failure vs route selection during normal operation.
Typically the latter is way important, obviously but it can be mitigated
somewhat by careful placement, topology-wise, of ones dedictated
route-reflector.

One other small point (and a potential flame attractor at that), route
selection in a reflected environment in and of itself can (and in a large
network, usually does) result in some levels of suboptimal route
selection. YMWV depending on your topology.

I would agree that 40 is a very nice number for a multi-use RR and that,
given a healthy quantity of memory in a dedicated RR, can probably be
doubled without significant additional convergance time.

-Steve

* Alan Hannan <alan@routingloop.com> [010501 21:40]: