Here is a preview of one of the slides I shall have
at NANOG.
This shows the number of flaps/second based bgp table
changes seen on four several SprintLink/ICM routers
(icm-dc-1.icp.net, sl-dc-8.sprintlink.net, sl-stk-5.sprintlink.net
and sl-mae-e.sprintlink.net). These are several of our busiest
boxes.
I took some care to get rid of strange effects caused by
router reloads and things I knew would have a measurable
effect on the results (clear ip bgp *, for example, or
turning off the line between sl-dc-8 and sl-mae-e). All
of these spiked in the 100-300/second range or in the
many-thousand/second range, so they tended to stand out a bit.
Matching reality to this output, routers sweat starting at
about the 14-15 mark, and at the 30/second mark, traffic is
noticeably affected. At the 50/second mark, these 7000s
will spend all their CPU on processing BGP updates.
Also of interest, I have three days' worth of scanned
output of debug ip bgp updates. I am crunching the
output now (it's *huge*)...
Anyway, this is a postscript thing, and it prints OK on
a QMS laser printer in landscape mode.
Sean.
- --
# This is a shell archive. Save it in a file, remove anything before
# this line, and then unpack it by entering "sh file". Note, it may
# create directories; files and directories will be owned by you and
# have default permissions.