Return to BSD News archive
Path: euryale.cc.adfa.oz.au!newshost.anu.edu.au!harbinger.cc.monash.edu.au!munnari.OZ.AU!news.mel.connect.com.au!news.mira.net.au!inquo!news.seinf.abb.se!nooft.abb.no!Norway.EU.net!EU.net!newsfeed.internetmci.com!in2.uu.net!news.mindspring.com!reflections.mindspring.com!tlewis
From: Todd Graham Lewis <tlewis@mindspring.com>
Newsgroups: comp.os.linux.networking,comp.unix.bsd.netbsd.misc,comp.unix.bsd.freebsd.misc
Subject: Re: TCP latency
Date: Tue, 9 Jul 1996 02:29:28 -0400
Organization: MindSpring Enterprises
Lines: 98
Message-ID: <Pine.LNX.3.91.960709020017.19115I-100000@reflections.mindspring.com>
References: <4paedl$4bm@engnews2.Eng.Sun.COM>
<31D2F0C6.167EB0E7@inuxs.att.com> <4rfkje$am5@linux.cs.Helsinki.FI>
<31DC8EBA.41C67EA6@dyson.iquest.net> <4rlf6i$c5f@linux.cs.Helsinki.FI>
<31DEA3A3.41C67EA6@dyson.iquest.net> <Du681x.2Gy@kroete2.freinet.de>
<31DFEB02.41C67EA6@dyson.iquest.net>
<4rpdtn$30b@symiserver2.symantec.com> <x7ohlq78wt.fsf@oberon.di.fc.ul.pt>
NNTP-Posting-Host: reflections.mindspring.com
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
In-Reply-To: <x7ohlq78wt.fsf@oberon.di.fc.ul.pt>
Return-Receipt-To: tlewis@mindspring.com
Xref: euryale.cc.adfa.oz.au comp.os.linux.networking:44439 comp.unix.bsd.netbsd.misc:3973 comp.unix.bsd.freebsd.misc:23099
Pedro, Terry, et al.,
I fly the flag of truce; please don't shoot.
On 8 Jul 1996, Pedro Roque Marques wrote:
> >>>>> "tedm" == tedm <tedm@agora.rdrop.com> writes:
>
> tedm> I feel this has gotten so academic that it is meaningless.
>
> I for one i'm so tired of seing non-techical arguments on Usenet about
> supposedly techincal issues that i don't ever consider a discussion to
> get "too academic".
Then let's get to it. A steak dinner at the next NANOG or LISA, my treat,
to anyone who makes a substantial contribution to this thread. There are
two questions in this message; answer them if you have the time.
> tedm> Who cares what the latency/throughput figures are or who is
> tedm> winning the current benchmark in vogue!
>
> There is two things here:
> I for one care about the latency/thoughtput figures since it helps to
> evaluate the code i intend to test and work on. I really don't care
> about who is winning but that is not to say that a comparison, when
> numbers are largely different is not useful.
Ditto. I am an unapologetic Linux user, and most of my coworkers are
FreeBSDers. No one wants a "My disk is bigger than yours" war, so let's
not have one.
> tedm> The fact is that these numbers only become important on
> tedm> servers that are being used to serve users, which strikes
> tedm> out 90% of the personal Unix boxes out there in my opinion.
Hogwash. sysctl is a reality under FreeBSD and shall soon be so under
Linux. Good networking code is a goal towards which all of us strive,
and it's not like we're hurting anyone by optimising.
> Having good thoughput and/or latency in TCP is much harder than most
> people believe.
Question #1: Which aspects of network performance under FreeBSD and/or
Linux are most in need of improvement? Extra credit for
well-reasoned answers. If latency is unimportant, which I
don't think anyone is seriously asserting, then what else is
important, to which everyone should have an answer.
> There is a full load of things going on in a TCP stack
> that are very dificult to deal with. If you want an precise example
> the BSD stack evolution from 4.2 to 4.4 has greatly improved those
> figures, the more important work on this, done over *years*, was done
> by the LBL people and specially Van Jacobson. Curiously enhough if you
> read his mails from the time you find several references to both
> thoughput and latency messured with unloaded machines.
These are obviously not cut and dried issues. TCP has been around
for 15 years and we're still wrestling with them. So please, enlighten
us end users, (and I mean that b/c I'd love to hear the various
opinions), what is left to be done?
> For someone that is working on the subject this numbers are very
> important.
(...)
> When to send an ack is not at all trivial by the TCP specs and
> influences bandwidth utilization and shows up on the latency test (if
> you send extra acks) or completly breaks your interactive performance
> (if you send too little or too delayed acks).
Question #2: Does latency optimization have ancillary benefits in terms
of general code robustness or quality?
> Now, can we have technical discussions between people regardless of
> their religious belief or choice of operating system ? There is a lot
> of stuff that i would like to learn with the BSD people... by acident
> i might even be hable to contribute something back to the pool.
Ditto on the former, although highly unlikely for the latter.
I really respect the FreeBSD guys. I think that you do very good work,
and I use your os every day with good results. As a good neighbor, I ask
you to do us a favor: Please don't storm off in a huff muttering about
"Those stupid Linux guys." If you have input to make on the network
optimization issue(s), then I would love to hear it.
So what is it? UDP performance as a function of packet size? Time of
context switch as a function of # of processes (as affects forking
network servers)? Multicast performance? Various network performance
measures as functions of size of routing table? OSPF or BGP
performance? NFS?
We're listening, and we're curious. (And we're offering steak dinners.)
_____________________________________________________________________
Todd Graham Lewis Core Engineering Mindspring Enterprises
tlewis@mindspring.com (Standard Disclaimers) (800) 719 4664, x2804