smd at icp.net
Mon Oct 23 05:05:54 UTC 1995
The information you supplied is more helpful for
figuring out what you're seeing than other things I have seen.
Indicating a traceroute and where you see packet loss,
and when you see packet loss is a big first step in
A couple quick points: the place to go to for this kind
of thing should be your direct service provider, who hopefully
has some means of communicating with upstream and peer providers
when things are going wrong somehow, or at least might be
able to give you some additional information.
NANOG and other lists aren't appropriate places to discuss
specific problems. OTOH, general problems, such as how big
the Internet is and how saturated well-known-things seem to
be and what might be done about it, could be OK here.
Your numbers are quite reasonable ones, and point to
heavy congestion in the area of MAE-EAST+.
A good part of that is on the DS3 between MCI and their
first MAE-EAST+ router. MCI has done something reasonable
about that, viz., putting a second router in place, which
I am about to set up a peering with. This should divert
a large amount of traffic away from the original DS3, which
in turn likely should improve performance between you and PSI.
I should also note that Sprint and MCI are in the final
phases of negotiating direct point-to-point peerings which
will pop up in several areas, and improve connectivity
between SprintLink and InternetMCI. You wouldn't be far off
in expecting this to become something of a trend,
particularly among a small number of the large heavy-traffic
providers, for whom the NAP/MAE/FIX concepts and models
don't seem to be scaling well in practice.
Meanwhile, expect congestion, particularly at NAPs/MAEs/etc. :(
| If you want a really good a measure, instead of ping try running NTP
| over the links in question. It gives a very accurate measurement!
| And I've observed 40% loss peaks there too!
NTP is one of two really useful network diagnostic tools,
and there is another by Matt Mathis that I hesitate to
advertise simply because it is somewhat CPU-unfriendly
towards busy Cisco routers, while the former is more useful
for long-term information gotten as a side-effect.
You might want to investigate the IPPM work, btw, wrt
Finally, wrt MAE-EAST, you might want to look at
http://www.mfsdatanet.com/MAE/east.stats.html for an
idea of the amount of traffic there. Hint: it's *lots*.
More information about the NANOG