On 7/09/2012 2:54 a.m., Daniel Beschorner wrote:
Congratulations on the 3.2 release!
It seems quite stable in our >1000 client environment, good work!
Some first impressions after using it for some days (Linux x64, no disk cache,
SMP mode with 2 workers):
On an old server (SLES 10) the load sharing between workers (2 or 4) is pretty
bad, in fact only one worker seems to do all the work. On a SLES 11 system load
sharing is a lot more fair. But as you stated is this OS related, not a direct
Squid issue.
The shared memory cache is harder to follow in its memory use than in 3.1, am I right that the
"cached" value of the "free" output is somehow related due to shared mem?
Where are you spotting this?
And some kind of little bug maybe related to SMP: If I rotate the logs (1x
access.log and 1x cache.log) they spread both to log.0/.1/.2 and suddently
access.log.0/access.log.1 and cache.log/cache.log.0 are used in parallel.
Ooh. Thank you for spotting that. Please register it as a bug.
However, 3.2 with SMP seems to be a really nice scalability feature.
Thank you!
Daniel
