From: alissitz@cisco.com
Date: Thu Sep 22 2005 - 12:20:58 GMT-3
If you can in their lab; simulate a neighbor going down (clear ip bgp on
a peer) but keep the link up. See what happens... If iBGP is peering via
loopbacks they may have a routing problem depending on how they have
configured static routes to the peer addresses... If they have
configured iBGP for next hop addresses, then bounce a link and see what
happens... Either routing will be disrupted or the change will not be
reflected throughout the network quickly. BGP is becoming a quicker
protocol with new features coming out, but IGPs have been this way for a
while now.
Do not use route dampening for financial customers... They live by the
data exchange and prolonging a route problem will cause any route
problem to be worse and last longer that it may have ... You do not need
route dampening for a small network anyway
If they insist on BGP, then look at fast failover and BFD to quicken
convergence / detection of a problem. If a neighbor bounces a lot ...
Then BGP graceful restart should be used ... All of these suggestions
have platform and code dependencies... It is no fun to have to check all
code levels and platforms that carry customer data when providing a
solution.
None of this is as pretty as typing "router eigrp/ospf x" ... --> after
typing this, go to sleep happy knowing that your IGP rocks! As Dave
said below, network speed is the key, speed / delay equals dollars
(sometimes millions) to these financials ... Stick with simplicity and
speed!
Andrew
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
Schulz, Dave
Sent: Thursday, September 22, 2005 10:34 AM
To: Tim ; nobody@groupstudy.com; 'Group Study'
Subject: RE: ibgp w/o IGP - How bad?
I would assume that this is the Stock Exchange, that fast convergence
may be the best selling point to look at. Speed/time is money in the
stock world.
Dave
-----Original Message-----
From: nobody@groupstudy.com
To: 'Group Study'
Sent: 9/22/2005 5:48 AM
Subject: ibgp w/o IGP - How bad?
Hi guys,
My client, a very well known Stock Exchange, is planning to re-engineer
his network.
The new network design calls for using iBGP between 4 fully meshed
peers, without any IGP.
Being a Stock Exchange, extreme reliability is paramount.
So far, I haven't been able to convince this client that this plan is
not a good idea.
I've given this client several reasons why this is not a good idea but
to no avail
Maybe I'm missing the most important and compelling reasons not to do
this.
I've told him that iBGP wasn't designed to be used in place of an IGP
because BGP's loop avoidance mechanism is based on the list of AS's
within the path attribute; within an iBGP mesh, there's no loop
avoidance mechanism.
I also told him that should any neighbor peering go down that
re-establishing that peering will be very slow (relative to IGP's).
IMHO, there are other very poor design choices, for example, a whole
bunch of static routes are being redist into bgp on one side of the iBGP
"cloud" and then being redist into RIP on the other side.
Besides the reasons I've already stated, are there any other MORE
important reasons this is a poor network design?
This design was tested and it does work but I'm concerned that any
problem - no matter how small - will cause the whole network to crash.
What do you guys think?
Tim
This archive was generated by hypermail 2.1.4 : Sun Oct 02 2005 - 14:40:16 GMT-3