Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 30 Sep 2003 17:19:10 +0200
From:      Pawel Malachowski <pawmal-posting@freebsd.lublin.pl>
To:        freebsd-ipfw@freebsd.org
Subject:   Re: Queue sizes.
Message-ID:  <20030930151910.GB22091@shellma.zin.lublin.pl>
In-Reply-To: <0AF1BBDF1218F14E9B4CCE414744E70F1F3F07@exchange.wanglobal.net>
References:  <0AF1BBDF1218F14E9B4CCE414744E70F1F3F07@exchange.wanglobal.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Sep 30, 2003 at 12:43:37AM +0200, Sten Daniel S?rsdal wrote:

> I've experimented with various queue sizes to pipes and i just cant figure out a generic algorithm from 10mbit through 64kbit
> (10240, 8192, 4096, 2048, 1024, 512, 256, 128, 64 kbit/s).
> Does anyone know the most efficient queue size for latency (most important) vs. bulk (must be _roughly_ same before and after) ? 
> Google doesnt help much (keywords might be wrong?).
> I would really appreciate if anyone got any tips/clues?

I've found for my own, that usually setting queue (buffer size, it is
a bit confusing to call `queue' two things) for about 1/3 of links
speed in KBytes is OK.

for example,

I set queue to 30KBytes for pipe with bw 768kbit/s.
When saturated, 768kbit/s (96KB/s) link can transmit 30KB within ~300ms
and it is acceptable *for me* (note, this 300ms must be usually doubled
because incoming and outgoing traffic have separate pipes).
Setting too small queue size can cause problems, for example TCP has
problems with packet lossess and data transfer can be lowered twice or
even more, I was observing this with WinXP system, this is normal so be
carefull and test a lot.


-- 
Paweł Małachowski



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030930151910.GB22091>