Ungodly packet loss rates

Mon Oct 21 22:12:39 UTC 1996

This whole thread really disturbs me.

I think the original message that started it was a bit out of line,
but only because it was cc-ed to NANOG without giving providers
a chance to respond to the criticism.

Nevertheless, the posts that followed it were far worse.

Ed Vielmetti have suggested interconnects are "to blame" and that if
you transit any of them you're likely to see loss, or that if you even
take multiple providers to get your destination, you might as well
throw in the towel.

Dalvenjah FoxFire complains about a similar amount of packet loss
across MAE-WEST. Unfortunately there aren't enough details to see if
the problem is related. But the implication is that MAE-WEST is at
fault. This is pretty unlikely, go look at the graphs on
http://www.mfsdatanet.com/MAE/west.stats.html, yourself. It is likely
that at any given point some provider may have congestion on one of
their links to an interconnect -- surely this does not mean the
interconnect was to blame.

Bill Bradford suggests you're lucky to get anything approximating
a lossless connection.

Derek Elder injects irrelevencies about holes in operating systems
that make us wish NANOG was a moderated list (too bad the only
qualified people don't have time to moderate).

To my mind, the right answer is to check with the folks involved!
Any other answer falls short. If there is loss somewhere, most
providers want to be told about it. In some cases, it's possible
to shift traffic around to deal with such problems. In other cases,
there may be planned upgrades in the works to deal with the problem.
In other cases, they may not even be aware of the problem.

To comdemn network providers without even asking them to justify and
investigate the loss represented by traceroute on their networks
is hardly fair. 

Oh yes, the second time around, Ed suggests that network operators may have "no
hesitation dropping traceroute or ping packets to low priority". No major
network provider *drops* traceroutes or ping packets. Many vendors' code (cisco
in particular) deprioritizes handling these packets such that round-trip-times
are not meaningful; said vendors also rate-limit responses -- this is normally
intrusive but might become so in the face of poor testing methodologies.

--jhawk
  John Hawkinson