Announcement Propagation Delay in BGP

Mailman List Mirror (Read Only)
1

I am going to be announcing two new prefixs into BGP soon
and the netgeek in me is very curious as to the length of
time it takes to show up in other parts of the world that're
logically far from Hawaii. Instead of going to
www.traceroute.org and refreshing repeatedly, I thought
folks here might've created a tool to do just that. Or,
perhaps, someone else has stats on a test they ran in the
past.

Anyone got any tools or suggestions on how to measure this
just for fun and not as a scientific study? I plan to read
the beacon paper "sahara.cs.berkeley.edu/papers/MMRB03b.pdf"
but I believe this mostly looked at how long-period flaps (2
hours) were still punished. Any other pointers?

Thanks,
scott

2

All your wishes come true: http://ris.ripe.net

Greets,
Jeroen

3

Scott Weeks wrote:

I am going to be announcing two new prefixs into BGP soon
and the netgeek in me is very curious as to the length of
time it takes to show up in other parts of the world that're
logically far from Hawaii. Instead of going to
www.traceroute.org and refreshing repeatedly, I thought
folks here might've created a tool to do just that. Or,
perhaps, someone else has stats on a test they ran in the
past.

Non-scientific test, but I've seen new prefixes appear at the Oregon-IX route server in <20 seconds (we're only in Texas), and reach a modestly steady state in <90 seconds. I've seen adjustments (prepending, etc.) appear in 45+ seconds, and withdraws in probably the same time.

pt

4

* Pete Templin:

Non-scientific test, but I've seen new prefixes appear at the Oregon-IX
route server in <20 seconds (we're only in Texas), and reach a modestly
steady state in <90 seconds. I've seen adjustments (prepending, etc.)
appear in 45+ seconds, and withdraws in probably the same time.

The is the mere propagation delay, and it's reasonable to expect it to
be in this range. Unless you botch something, the route flaps a
couple of times, and the prefix is dampened. 8->

There might be some delays because prefix filters aren't updated in
real time, even if they are automatically generated from RPSL
snippets. I don't know if this makes a difference in the real world,
though.

If IANA assigned the surrounding /8 to the RIR just a couple of weeks
ago, you might run into some static access or prefix lists.

To get a global picture, I think you'd have to ask the GRADUS folks.
RIPE RIS is interesting, too, of course.