Subject: Re: statistics on retransmitted packets
To: Laine Stump <lainestump@rcn.com>
From: Greg A. Woods <woods@weird.com>
List: netbsd-users
Date: 11/15/2000 18:20:23
[ On , November 11, 2000 at 17:24:05 (-0500), Laine Stump wrote: ]
> Subject: statistics on retransmitted packets
>
> Is there a simple way to get statistics out of IP about the percentage
> of retransmitted TCP packets over a period of time? My cable line (RCN
> in the Boston area) is getting progressively worse, with RTTs to the
> next hop often > 500ms, ping lossage commonly above 50%, and frequent
> total outages of several minutes

I know exactly what you mean!  This summer I suffered similar problems,
and they're not completely gone yet.

My current provider is Rogers@Home, using a Terayon modem here in
Toronto.

I've been using Eric Wassenaar's somewhat superior version of ping
(<URL:ftp://ftp.nikhef.nl/pub/network/ping.tar.Z>), combined with
Cricket (<URL:http://cricket.sourceforge.net/>), to get a rough measure
of packet loss over my cable modem link.

You can find my Cricket setup posted to the cricket-users mailing list
(<URL:http://www.geocrawler.com/archives/3/1089/2000/6/0/3965267/>, and
<URL:http://www.geocrawler.com/archives/3/1089/2000/10/0/4500259/>).

Occasionally when things get really bad (i.e. when the averages go over
10% during peak usage times) I send a copy of my Cricket graph image to
<support@rogers.home.net> with a big nastily worded complaint and
usually a day or so later things improve a bit again.

<URL:http://becoming.weird.com/~cricket/grapher.cgi?target=%2Fnet-rtt>

As you can see from my graphs my broadband link into their network is
somewhat better recently, but there's still tons of loss on their
internal backbone between routers.  There are also still lots of
problems with modem and/or access link resets giving short-term (1-2
min) outages.  My worst problems were during July and August, though
even through Sept. there were still significant problems.

Rumour has it that Rogers sold *WAY* more hookups this summer than they
had infrastructure to support and now they're begging to get enough
head-end gear and whatnot to make things even out again.  It would seem
they've now got enough bandwidth for my gateway, but I still can't get
to some other parts of their network without seeing lots of loss (and
that often includes their main outbound Internet connection).

Ted Rogers, the big boss @ Rogers, tells the media that users should
expect some teething problems with such a wonderful new service, but the
way these greedy cable monoplies over-sell services they don't
understand is rather annoying to those of us who pretend to know more
than they do.

BTW, if you are using a Terayon modem, and you are pinging your default
router, which is the Terayon "gateway" machine (which usually fronts a
number of "link" machines), then there's no loss due to echo requests
being handled at a lower priority -- i.e. in that case they're either
mis-informed, lying, or both.  So far as I can tell both from doing
experiments with ping floods under load (I do high-level support for a
small broadband provider using the Teryaon gear), and from what I can
glean from the local Terayon tech support folks, the gateway machine
always answers all pings at full priority -- at least nobody has
officially denied my claim.  The most curious part about the Teryaon
gear is its ability to hold onto an echo request and respond *many*
seconds later (eg. up to 30s!).  Also curious is its ability to send
duplicate replies.

Rogers also still uses lots of the LanCity gear too, as does the
provider I help out, and that's the type of modem my network tunnel
terminates on.  The LanCity gear seems to have different responses to
heavy loads and it doesn't seem to suffer loss as much as just slow
down.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>