[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Ping times & traceroute divergence



>>>>> "sg" == Steve Glines <sglines@is-cs.com> writes:
> At the moment I traceroute to Brazill but here are my ping times:
> To Brazill (aroot.psg.com)
> round-trip min/avg/max = 519.7/524.3/528.9 ms
>
> To  mbone-cern.switch.ch
> round-trip min/avg/max = 149.9/149.9/149.9 ms
>
> Interesting! [...]

Most of the Internet sees a shorter AS path towards the server in
Brazil (behind AS1251) than to the one in Switzerland (AS559).  For
example, this is a set of paths seen on the Oregon-IX route viewer:

with AS1251:

1740 7018 3561 1251 4128
2497 3561 1251 4128
8517 9000 2548 3561 1251 4128
7018 3561 1251 4128
2914 3561 1251 4128
2685 3561 1251 4128
715 3561 1251 4128
3561 1251 4128
2551 3561 1251 4128
2914 3561 1251 4128

with AS559:

3333 1103 8933 559 4128
1755 3300 3300 3300 3300 8933 559 4128
1 3303 3303 559 4128
2828 3303 3303 559 4128
10764 1 3303 3303 559 4128
3557 6461 3303 3303 559 4128
3557 6461 3303 3303 559 4128
5459 786 8933 559 4128

AS1251 is directly connected to a "Tier 1" provider (AS3561), while
AS559 connects to "Tier 1" providers (AS1, AS174, AS3549, AS3561,
AS6461) through an intermediate transit provider (AS3303).  In
addition, AS3303 prepends its own AS, probably for load balancing
reasons.

In general, having a long AS path towards the (US) "backbone" is
beneficial for a European provider, because it makes it less likely
that other European networks route to us through the US if alternative
routes within Europe are available.  (Maybe this is not so much of an
issue in Brazil---otherwise they should prepend AS1251 when announcing
to AS3561.)  So the route to Brazil looks shorter than the one to
Switzerland, even for many networks within Europe.

Even if, say, Randy injected AS4128 behind AS3130 (one hop behind
Verio's AS2914), many networks in North America would still find
Brazil closer (if they have a shorter path to C&W than to Verio).

Maybe we shouldn't take this too seriously as a problem, because the
current population of AS4128 injectors is far too sparse to expect
optimal routing from anywhere except AS559 and AS1251 and maybe their
immediate neighbors.
-- 
Simon.