Re: [squid-users] squid freebsd aufs + coss - same hd

From: Amos Jeffries <squid3@dont-contact.us>
Date: Fri, 7 Dec 2007 12:19:58 +1300 (NZDT)

> and losing the cache swap log just triggers a rebuild. You won't lose
> the cache, just time.
>
> Yes you have to set the swap state directories when using COSS.
>

Ah. darn COSS. We need to get rid of that handicap sometime :-(

Amos

>
> Adrian
>
> On Fri, Dec 07, 2007, Amos Jeffries wrote:
>> > rebuild of coss take longer time... about 30 mins to rebuild 6gb (with
>> > configs posted previously)...
>> >
>> > correct ?
>> >
>> > i will loose my cached files (aufs) if cache_swap_log is set to the
>> > same location of old swap file ?!
>>
>> 1) cache_swap_log is obsolete. Its now cache_swap_state to better
>> reflect
>> the file to which it applies, and that the file is NOT a normal log
>> file.
>>
>> 2) Better leave the cache_swap_log/cache_swap_state at defaults (missing
>> from squid.conf) for squid to handle it safely.
>>
>> 3) What Adrian meant was the order of the lines in squid.conf. Not an
>> actual re-ordering of cache-dir. This is a major reason for #2, so a
>> re-ordering in the config does not screw up the custom state file
>> numbering.
>>
>>
>> Amos
>>
>> >
>> > thanks !
>> >
>> > regards
>> >
>> > Alexandre
>> >
>> > On Dec 6, 2007 8:50 AM, Adrian Chadd <adrian@creative.net.au> wrote:
>> >> cache_dir's are 'checked' in order; I suggest putting the coss
>> >> directories
>> >> first.
>> >>
>> >>
>> >>
>> >> adrian
>> >>
>> >>
>> >> On Fri, Dec 07, 2007, Amos Jeffries wrote:
>> >> > Alexandre Correa wrote:
>> >> > >can 2 cache_dir on the same hd (dedicated for squid) cause
>> >> performance
>> >> > >impact ?
>> >> > >
>> >> > >i?m using cache_dir aufs ... and cache_dir coss (objetcs smaller
>> than
>> >> > >1000k)
>> >> > >
>> >> > >cache_dir aufs /var/spool/squid/cache 110000 32 256
>> >> > >cache_dir coss /var/spool/squid/onda_coss01 6000 max-size=1000000
>> >> > >maxfullbufs=4 membufs=20 block-size=4096
>> >> > >
>> >> > >
>> >> > >thanks !!
>> >> > >
>> >> >
>> >> > Looks good. Just some hints though...
>> >> >
>> >> > You'd do well to set min-size on the AUFS dir, to push the smaller
>> >> > objects into COSS. At present small objects can go to either, and
>> >> large
>> >> > only in AUFS.
>> >> >
>> >> > A 4KB block-size may cause a lot of watse if you get a large number
>> of
>> >> > small objects such as optimized web pages or spacer images. The
>> >> default
>> >> > 512 is sufficient for COSS dir up to 8GB large
>> >> > Not worth changing it now though if any important data has
>> >> already
>> >> > gone to COSS. Would require a destroy and rebuild to fix
>> that.
>> >> >
>> >> > Two on same HDD might drag each other down a little, but no more
>> than
>> >> a
>> >> > single large cachedir doing the same throughput. Unfortunately
>> squid
>> >> is
>> >> > not head-optimised for disk usage yet. COSS is more in-memory than
>> the
>> >> > others so it should still be a net gain over a single pure aufs.
>> >> >
>> >> > Amos
>> >>
>> >> --
>> >> - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid
>> >> Support -
>> >> - $25/pm entry-level VPSes w/ capped bandwidth charges available in
>> WA -
>> >>
>> >
>> >
>> >
>> > --
>> >
>> > Sds.
>> > Alexandre J. Correa
>> > Onda Internet / OPinguim.net
>> > http://www.ondainternet.com.br
>> > http://www.opinguim.net
>> >
>>
>
> --
> - Xenion - http://www.xenion.com.au/ - VPS Hosting - Commercial Squid
> Support -
> - $25/pm entry-level VPSes w/ capped bandwidth charges available in WA -
>
Received on Thu Dec 06 2007 - 16:20:17 MST

This archive was generated by hypermail pre-2.1.9 : Tue Jan 01 2008 - 12:00:01 MST