More stuff derived from the several days' worth of debug
ip bgp updates...
There are three slides in the shar archive below:
-- S.ps is the first; that shows the up transitions
per hour for the top twenty most flap-propagating
neighbours of AS 1239. SPLK is SprintLink,
PSIAGG is an as-set due to proxy-aggregation done
on some of PSI's netblocks. The as-set is
{174,1800,690,2149}.
You will quickly see why there are three slides
instead of one...
-- T.ps is the second, and shows AS 1800 decomposed.
-- U.ps is the third, and shows AS 690 decomposed.
U.ps is intersting; it shows that PSI is flapping
*alot* at AS 690. This could be due to MAE-EAST
flakiness. It also shows that AS 690 propagates
alot of flaps from AS 3561. I don't know why.
PSI, ANS, were your peering sessions flapping
badly at MAE-EAST earlier this week?
Total number of flaps/hour in these graphs and in the
previous ones derived from this data set is 6309 (1.75/hour).
(The mean number of bgp table changes/hour in my first graph
is 4.32/hour. Math was done on a Pentium. 
Oh, and the
data sets covered different periods and had different
points-of-view.)
As I said, Sprint propagates all sorts of flaps all over
the place...
At NANOG I shall be talking about ways each of us can
help reduce the overall number of flaps.
What is still crunching now (and hopefully will finish
soonishly... i/o is slow) is an attempt to track intervals
betweens up and down transitions for any given prefix, the
number of prefixes that flap in any given transition, and a
very ugly attempt to show how much CIDRizing blocks of
prefixes could help).
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.