[22652] in North American Network Operators' Group
Re: latency vs. packet loss
daemon@ATHENA.MIT.EDU (Jon Zeeff)
Thu Jan 14 09:15:15 1999
To: amb@gxn.net (Alex Bligh)
Date: Thu, 14 Jan 1999 08:57:06 -0500 (EST)
Cc: nanog@merit.edu
In-Reply-To: <199901141105.LAA10096@diamond.xara.net> from "Alex Bligh" at Jan 14, 99 11:05:30 am
From: jzeeff@whs.verio.net (Jon Zeeff)
Reply-To: jon@zeeff.com
True, one can certainly think of and even find many situations where
they don't correlate, but in real world measurements, it looks like in
perhaps 90% of the cases, packet loss between randomly choosen places
on the Internet is accompanied by greater than typical latency. I suppose
this suggests that saturated links (where the router/switch adds latency by
buffering) are a commmon cause of Internet packet loss (vs line
errors, etc which would not show this correlation).
As a single example, connectivity from here to a popular NSP web site
is 40 msec in the morning with no packet loss and 500 msec with
20% packet loss in the afternoon. Routing is the same (through mae-east :-)).
> > How well does latency correlate to packet loss on the Internet? For
> > example, if one were to pick one of several randomly placed sites on
> > the net based on lowest latency to/from point x, what percentage of
> > this time would this also yield the site with the lowest packet loss
> > to/from point x? My guess is that the correlation is high (due to
> > typical buffer sizes).
>
> Remember latency is also affected by other things, like distance
> (you won't get less than 70ms RTT NY<->Lon even on an empty STM-1),
>
> Also note there are some conditions which cause packet loss which
> won't cause ICMP latency (line errors, various IXP overload conditions
