Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 4 Feb 1999 08:50:42 -0600
From:      Guy Helmer <ghelmer@scl.ameslab.gov>
To:        "Dirk-Willem van Gulik (vaio)" <dirkx@webweaving.org>
Cc:        freebsd-hackers@FreeBSD.ORG
Subject:   Re: Irratic Curve
Message-ID:  <Pine.SGI.4.05.9902040836570.2662-100000@demios.scl.ameslab.gov>
In-Reply-To: <36B97B38.74FF0B68@webweaving.org>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 4 Feb 1999, Dirk-Willem van Gulik (vaio) wrote:

> Whilst playing with a small, but fast, berkely DB based transaction
> server; which sits on a tcp/ip socket connection I ran into sometimes
> unpredictable reply times. One of the major problem was solved by 
> increasing the MSIZE to 256 (the 103 bytes+ delayed ack problem).

A performance fix for message sizes around 100 bytes was committed and
will appear in the 3.1 release.

> Now recently I came across:
> 
> http://www.scl.ameslab.gov/Projects/Gigabit/performance/prelim.html
> 
> Now could any one explain to me WHY freebsd appears so unpredicatable ?
> i.e. not a nice S-curve ? Is that the way of measuring ? Some other
> artifact, or real ? I think it is real, as I get the same sort of
> holes in my graphs for the transaction server.

Well, if you look at figure 3 on that page, it shows how changing the TCP
stack to not use delayed acknowledgements eliminates the dropouts (I
noticed that the change to the TCP stack is not explained on the page you
reference, but it is briefly mentioned at
http://www.scl.ameslab.gov/Publications/Gigabit/tr5126.html).

You can take advantage of a similar change to the FreeBSD TCP stack in
versions 3.0 and later by doing "sysctl -w net.inet.tcp.delayed_ack=0" to
turn off delayed acknowledgements.  I believe this results in
non-RFC-compliant behavior (I'm sure someone more familiar with the
standard RFCs will correct me if I'm wrong), but it does improve
performance dramatically in some circumstances.

Guy

Guy Helmer, Ph.D. Candidate, Iowa State University Dept. of Computer Science 
Research Assistant, Ames Laboratory       ---         ghelmer@scl.ameslab.gov
Research Assistant, Dept. of Computer Science   ---   ghelmer@cs.iastate.edu
http://www.cs.iastate.edu/~ghelmer



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.SGI.4.05.9902040836570.2662-100000>