[ previous ] [ next ] [ threads ]
 
 From:  "Holger Bauer" <Holger dot Bauer at citec dash ag dot de>
 To:  <m0n0wall dash dev at lists dot m0n0 dot ch>
 Subject:  M0n0 1.3b2 userland freeze on high load
 Date:  Wed, 31 Jan 2007 08:36:43 +0100
Hi,

Yesterday I have found out that m0n0 sees some similiar problems with
freebsd 6.2 that we encounter with pfSense as well. Under high load
userland processes freeze. I did some tests yesterday with a bridging
setup (OPT1 bridged to LAN) to simulate some delay for a
VOIP-Application (customer wanted to "hear" a 120 ms delay) so I
configured the trafficshaper to add it. As I had another setup handy I
flooded m0n0 running on a WRAP with 40 concurrent VOIP-calls on g711.
Though the bandwidth is rather low for this the pps-rate is rather high.
When doing the tests the webgui was not accessable anymore. Also the
svg-trafficgraph stopped working until the test was over.

We are seeing the exact same problem with pfSense. Enabling polling
works around that condition but nevertheless such a situation should not
occur imo. We already tried to get this problem on the appropriate
freebsd lists but the more or less general answer only was "your
hardware is too slow then". Maybe anyone here on the list has a clue
what to try to avoid this freeze?

Another thing that I noticed when doing some tests is that m0n0 now
performs at the same level like pfSense (though lots of things are
different like the used filter) when it comes to throughput on freebsd
6.2. I wonder if somone knows some tweaks that we have not already tried
;-) (though we tried hard to squeeze out extra mb/s already).

This discussion should help both projects to improve performance on low
end platforms so hopefully you don't blame me for posting this here ;-)

Holger