[Outages-discussion] [outages] AT&T SFO to Twitter -- possible routing issue or BGP hijack

Jeremy Chadwick jdc at koitsu.org
Sat Jun 24 03:58:32 EDT 2023


On Sat, Jun 24, 2023 at 07:30:45AM +0000, Jeremy Chadwick via Outages-discussion wrote:
> On Sat, Jun 24, 2023 at 06:55:37AM +0000, Job Snijders via Outages wrote:
> > On Sat, Jun 24, 2023 at 06:17:12AM +0000, Jeremy Chadwick via Outages wrote:
> > > $ telnet route-server.ip.att.net 80
> > > Trying 12.0.1.28...
> > > telnet: connect to address 12.0.1.28: Connection refused
> > > telnet: Unable to connect to remote host
> > 
> > It's a telnet service:
> > 
> > $ telnet route-server.ip.att.net
> > Trying 12.0.1.28...
> > Connected to route-server.ip.att.net.
> > Escape character is '^]'.
> > -------------- route-server.ip.att.net ---------------
> > ---------  AT&T IP Services Route Monitor  -----------
> 
> (Moving this to outages-discussion to keep from potentially paging
> more people.  ;) )
> 
> Thanks Job!  Someone should probably update PeeringDB to not specify
> http:// as the URI scheme then, or alternately use telnet://.
> 
> Anyway, back to the issue at hand:
> {snip}
> Gut feeling right now says PCCW Global may have some suboptimal routing.
> Doesn't seem to make sense why these packets would go all the way to
> Osaka.  Might be useful to see what PCCW Global is advertising.  And
> naturally, I can't see what the return path looks like.

Maybe this is not so much a "routing issue" as much as it could be an
anycast issue.  I suspect Twitter uses anycast, since trying to reach
104.244.42.193 from a system I have access to in France results in a 9ms
response time (staying locally within France per mtr), while from my
AT&T Fibre connection things go all the way to Osaka.

Same happens if I flush my DNS resolver/nameserver caches resulting in
www.twitter.com resolving to 104.244.42.65 -- AT&T Fibre goes via PCCW
Global in Osaka, while aforementioned host in France stays in France at
9-10ms.

Not that Twitter is particularly important to me (it isn't), but I found
this performance to be pretty sub-optimal and odd.  And obviously I have
no familiarity with Twitter's peering arrangements.

The connectivity issue that originally lead me down this path is over
with regardless.

-- 
| Jeremy Chadwick                              jdc_at_koitsu.org |
| UNIX Systems Administrator                      PGP 0x2A389531 |
| Making life hard for others since 1977.                        |



More information about the Outages-discussion mailing list