Return to BSD News archive
Path: sserve!newshost.anu.edu.au!harbinger.cc.monash.edu.au!msunews!news.gmi.edu!zombie.ncsc.mil!news.mathworks.com!europa.chnt.gtegsc.com!salliemae!uunet!news.thepoint.net!news.thepoint.net!not-for-mail
From: arlie@news.thepoint.net (Arlie Davis)
Newsgroups: comp.unix.bsd.freebsd.misc
Subject: Help! My FreeBSD TCP stack is broken?
Date: 5 May 1995 17:47:15 -0400
Organization: ThePoint - Kentuckiana's Internet Connection
Lines: 67
Message-ID: <3oe6d3$1mg@dg.thepoint.net>
NNTP-Posting-Host: dg.thepoint.net
I've been using FreeBSD for a month or two now, and I'm very pleased with it.
I've replaced every Linux machine in my network with FreeBSD, and haven't
regretted it -- yet.
I just installed FreeBSD on a machine today, and for some odd reason,
TCP connections are working _very_ slowly, but only to other FreeBSD machines.
My network looks like this:
Core network (Ethernet)
|
+--- FreeBSD machine (mbone), serving as mrouter
|
+-----ISDN----- My home, with two FreeBSD boxes (atlas and diesel)
|
+-----ISDN----- Friend's home, with one FreeBSD box ("firewall")
|
+---28.8K modem--- Linux box (router) ---- Ethernet
|
(Remote network)
|
FreeBSD box ("charon")
Most of the BSD machines work beautifully. mbone, atlas, diesel, and
firewall all intercommunicate as expected. (All are involved in multicast
applications, but that's not what this is about.)
The last machine, charon, has BIG problems. From what I can tell, _any_
TCP connection between charon and any other FreeBSD machine has the problem.
The problem is that TCP sessions move very, very slowly. Between a keypress
and the echo of the character, there is a delay of more than 5 seconds.
(The total latency of the link is only around 240ms.) Note that ICMP pings
act as expected -- close to the link latency.
_Strangely_ enough, charon has no trouble with TCP conversations with other
machines.
I looked at all the modifiable options I could find on the TCP. I tried
disabling T/TCP on both ends, with no effect.
One clue: I tried taking the Linux box next to charon out of the picture
by using the PPP software on charon to directly connect to my terminal
server at my core network. This worked. And when charon was connected
in this way, all TCP sessions work as expected! Oddly enough, though,
_no_ other machine that uses the Linux box for a router has any problem
with it at all -- it has performed well.
This points to a problem with my remote Ethernet network. But why
would it only affect TCP between FreeBSD machines??
Note: The machine charon is a simple, low-end workstation: 486 2/66,
8M RAM, IDE disk (408M). Both charon and the Linux router are using
3c509s on a 10Base2 thinnet Ethernet.
If anyone has _any_ clues about this behavior, please contact me. I'm
surprised to see this behavior from an otherwise solid product.
Please mail me at <arlie@thepoint.net>. Thanks for any advice.
--
-- Arlie Davis | The Point: High-quality Internet service. 82G online.
-- <arlie@thepoint.net> | $20/mon (flat) (812)246-8032 http://www.thepoint.net
-- System administrator | -----------------------------------------------------
-- E Pluribus UNIX | # ./programmer < coffee > main.c