Re: squid2.4Devel2 - CPU Usage + Various error messages

On Thu, 29 Jun 2000, Vincent Y H Man wrote:

> Hello all,
>
> Newbie squid user here. Using a development version because I wanted to
> see if I could recreate similar results to Duane's fine showing at the
> bakeoff (congrats).
>
> Im Using polygraph 2.2.9 to benchmark squid 2.4 with FreeBSD 3.3
> as my OS. This is a dedicated squid server. It's not running any other
> services.
>
> Hardware:
> PIII 550 MHz
> 396MB Ram (128MB cache_mem)
> 2*8.5 scsi HDs (cache_dir diskd 3000 16 256 * 2) <-- system files are on
> one of these drives (because I don't have an IDE handy =P)
> 3com Fast ethernet @ 100Mb
>
> Here's the problem:
> At peak load levels during the test (~90 req/s), squid decides to do a
> "StoreDirWriteCleanLogs". This ties up system resources and causes squid
> to respond much slower to requests.

> Questions:
> What triggers this operation?
> Is there any way to avoid this tying up of system resources?

StoreDirWriteCleanLogs should happen in one of two cases:

        1) someone (i.e. cron) runs 'squid -krotate'

        2) squid traps a fatal error and tries to save the cache logs
           before dying.

> Unrelated problem:
> Error Messages:
> /kernel:xl0:command never completed
> /kernel:xl0:transmission error:90
> /kernel:xl0:tx underrun, increasing tx start threshold to 180bytes
>
> Never seen these messages while running Squid 2.3 Stable3. Same NIC,
> different LAN speed (connected at 10Mb before)

I've had bad luck with hardware that uses the xl0 driver. Or maybe
its the driver itself. I use Intel NIC (fxp0) with great success
(except on Linux).

Duane W.
Received on Thu Jun 29 2000 - 15:45:07 MDT