Re: Squid Error Question

Yongmin LEE wrote:

> 1999/07/21 17:37:22| comm_poll_incoming: NULL write handler
> 1999/07/21 17:37:23| comm_poll_incoming: NULL write handler

You probably have a ICP peer which is refusing ICP queries (not
listening on the ICP port set in cache_peer).

> 1999/07/21 18:01:05| aio_queue_request: WARNING - Running out of I/O theads
> 1999/07/21 18:01:05| aio_queue_request: Perhaps you should increase NUMTHREADS
> 1999/07/21 18:01:05| aio_queue_request: Or install more disks to share the load
>
> How to fix this error?

If the disk is busy all the time then you may need more drives to handle
the load, or use max_open_disk_fds to limit the load Squid puts on the
drives (will also lessen your hit rate when load is high).

If the disk is not very busy then you need to increase NUMTHREADS in
src/aiops.c and recompile Squid. You may also be interested in a
async-io performance patch of mine to cut down on the number of I/O
threads and CPU used by Squids async-io code. The patch is available
from my Squid page <http://hem.passagen.se/hno/squid/>.

--
Henrik Nordstrom
Spare time Squid hacker