Hello,
Are there any known issues with Linux RedHat 6.1 and Squid 2.4STABLE1? I have been running Squid 2.3STABLE1 through STABLE5 for many months without problems, however whenever I try to implement 2.4STABLE1 to benefit from AsyncIO or DISKD, as soon as the server has finished validating the store, the load average jumps through the roof (20 or more!!!). When AsyncIO is configured, within 30 seconds I get warning messages of IO overloading. Please note that the above is observed even though I have not yet begun to send requests to the cache!!!!! The machine basically comes to a crawl almost instantaneously with no client requests. When using DISKD, load average sky-rockets and tests of sending only a few client requests have 6000ms or more before a response is given with less than 60 reqs/min.
I am using the following configuration directives:
--enable-snmp --disable-poll --disable-ident-lookups --enable-delay-pools --enable-underscores --enable-removal-policies=heap --enable-async-io=64 --enable-storeio=aufs
The machine specs are as follows:
PentiumIII 800Mhz
Linux 2.2.19 with 8192 fds available
glibc-2.1.3
egcs-2.91.66 19990314/Linux (egcs-1.1.2 release)
768 MB RAM
3 x 18G 10K RPM SCSI drives
aic7892: Ultra160 Wide Channel A, SCSI Id=7, 32/255 SCBs (this is the adaptec driver)
I tried using --enable-poll and --disable-poll, and so far even on 2.3STABLE5 using --disable-poll seems to give a more responsive cache.
Any help would be appreciated, as I have been looking forward to using AsyncIO for a while but have yet to be able to!
Thanks,
Ron