[ previous ] [ next ] [ threads ]
 From:  "Seth Martin" <SethM at turbinegenerator dot com>
 To:  "Darryl Okahata" <darrylo at soco dot agilent dot com>
 Cc:  <m0n0wall at lists dot m0n0 dot ch>
 Subject:  RE: [m0n0wall] Traffic Shaper Latency Take 2
 Date:  Wed, 24 Aug 2005 12:37:02 -0400
Sorry if I wasn't clear, I get latency in either direction as long as
it's under some sort of load with traffic shaping enabled.  I can
disable the traffic shaper and upload or download at the full 1.54 no
problem.  Under full load with no shaping I gain about 50ms of latency,
with the traffic shaper I gain about 300-500ms depending probably on the
size and type of packets.  

I just checked the load and its peaking upwards towards 50% usage under
full load with the shaper on, and without it it barely budges.  I'll try
the P4 this week.  Maybe the extra cache and bus will help it route the
traffic faster.

-----Original Message-----
From: Darryl Okahata [mailto:darrylo at soco dot agilent dot com] 
Sent: Wednesday, August 24, 2005 11:39 AM
To: Seth Martin
Cc: m0n0wall at lists dot m0n0 dot ch
Subject: Re: [m0n0wall] Traffic Shaper Latency Take 2 

"Seth Martin" <SethM at turbinegenerator dot com> wrote:

> I have tried and tested things with the traffic shaper, and with just
> two 500k pipes, one in and one out, I get latency under full load in
> one direction, with 1 1000k pipe for both directions I get latency.

     Have you actually tested the bandwidth in both directions (without
any shaping)?  The fact that you get latency in only one direction is
odd (I assume that you're testing for full load in both directions, but
only getting latency in one).

[ Also, with 1.11 at least, I seem to recall that the setup fields are a
  bit confusing: bits per sec vs bytes per sec (I don't think the fields
  say which units to use).  Get them wrong, and you'll see latency.
  IIRC, you also need to set the values just slightly below the "actual"
  values.  ]

> Could this be a limitation of the soekris 4801's processor, would
> installing it in a spare p4 system with some intel nics fix the
> problem?

     You might want to periodically poll the Soekris's load average via
the exec page.  That might be a simple way of telling you if the Soekris
is getting hammered or not.

	Darryl Okahata
	darrylo at soco dot agilent dot com

DISCLAIMER: this message is the author's personal opinion and does not
constitute the support, opinion, or policy of Agilent Technologies, or
of the little green men that have been following him all day.