[squid-users] squid cache Large rock with aufs optimization for bandwidth saving

Amos Jeffries squid3 at treenet.co.nz
Sun Nov 16 20:19:17 UTC 2014


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 17/11/2014 1:26 p.m., Ahmed Allzaeem wrote:
> Hi Amos , thanks for reply.
> 
> But I think using large rock will let the memory cache size be >32
> ??? am  I correct ?
> 

Maybe no, maybe yes. Shared memory is a separate feature to large
rock. They previously had the same size limitations. largerock solved
it for rock caches. But the documentation does not show any change for
shared memory.


> I mean although im not using rock store , but I was thinking will
> have memory size >32Kb
> 
> So , I keep the aufs size which I assume will help me and will let
> me break the 32kB limitation .. but it seems not like I was
> planning !!!!
> 
> So m my previous config was using aufs with large rock needed in
> memory size which is not what I was planning ?!!!! can you plz
> explain me why my previous config don’t have luck with memory >>
> 32KB??
> 

Explained that in tha last email. AUFS is per-worker caching and
memory_cache_shared probably still has the 32KB limit.

> 
> So for now , as u told me I removed all aufs drivers and added only
> rock dir as below :
> 
> cache_dir rock /ssd1 90000 max-size=32768 swap-timeout=350
> max-swap-rate=350 cache_dir rock /ssd2 90000 max-size=32768
> swap-timeout=350 max-swap-rate=350 cache_dir rock /ssd3 90000
> max-size=32768 swap-timeout=350 max-swap-rate=350 cache_dir rock
> /ssd4 30000 max-size=32768 swap-timeout=350 max-swap-rate=350
> 
> 
> And only mean size I can see now is 18 Kb ??!!!!

This has manually limited the rock dirs to ~32KB (max-size=32768).

> 
> I had tried another trial , I put memory_cache_shared off , squid
> cant be started all all kids is being killed and stared again  and
> so on !
> 
> Sample of logs : FATAL: Received Segment Violation...dying.

Do you have gdb?

http://wiki.squid-cache.org/SquidFaq/BugReporting has details on how
to use it to get a backtrace we will need to fix this bug.

Amos
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.22 (MingW32)

iQEcBAEBAgAGBQJUaQbFAAoJELJo5wb/XPRj5JQH/03JULTbnGNgX5gz3wEw00eU
oTfcC/zJlwvItiuw/aOSsPoJ0zm9nTKjT/2BLcdFSS0hlk9gOczH0ea9OPAxDRvx
1u3MDNMcINAXo/YB8MXTgqp1fxS30OOkzILv0eoIQ+RJ/xtdIAJdvpxSy5DGHZnh
9b3VFXafAimOt+ZkWfML4b16Q5QAnHX1nHgWb9NHvubUtiB3B5eYw8+nHIOPbEtK
bVZrqVKQkxVbP3Bp+9NFrCrE6uhFw/Z9Xudc/hfBTec/8Moo/cw9pV1K0kqbzx8G
Vy06PI0RZBWRMZyBsXDY22aTrsslKtMyfbehuPCoVRbEETxY3jv8itntOT+kb7E=
=Vmca
-----END PGP SIGNATURE-----


More information about the squid-users mailing list