Re: [squid-users] Mem Cache flush

From: Manoj_Rajkarnikar <manoj@dont-contact.us>
Date: Sun, 20 Jan 2008 11:57:38 +0545 (NPT)

Hi Haytham,

On Fri, 18 Jan 2008, Haytham KHOUJA wrote:

> Hello,
> I don't advise you to use partitions as separate cache_dirs. You will not
> get performance enhancement since you're still working on the same physical
> disk and same SCSI controller.

Its not 2 partitions on 1 physical disk, I have 2 scsi disks... controller
however is same.

> Moreover, 1GB for cache_mem is way to much if you only have 3GB of total
> RAMs if you have 8x8GB cache_dir.
> Lower that to around 256 MB ~ 512 MB and try.

Let me try that... I still have free mem though. by the rule of thumb, we
need 10MB Ram for indexing per GB of cache. that sums up to
8 x 8GB x 10MB/GB = ~650MB RAM for cache index + 1GB for Cache_mem =
~1.7GB. so I'd still have ~1.3GB of RAM for system operation.. I'll lower
the cache_mem, however, to try..

Thanks.

Manoj

>> Hi all.
>>
>> I'm having an issue with our cache. its a 2.6S17 on centOS 4 x86_64 with
>> 3GB Ram 2x36GB scsi disk for cache_dir and separate SATA 80GB for system.
>> It seem to flush the mem cache for no apparent reason.. nothing in the
>> cache log around the time of occurance... please see the graph in the
>> following link.. it has done it twice in 10 days. we've not encountered
>> such issue.
>>
>> relative squid.conf settings follows:
>>
>> cache_mem 1024 MB
>> maximum_object_size_in_memory 16 KB
>> cache_replacement_policy heap LRU
>>
>> cache_dir aufs /sda1/cache0 8000 20 256
>> cache_dir aufs /sda1/cache1 8000 20 256
>> cache_dir aufs /sda1/cache2 8000 20 256
>> cache_dir aufs /sda1/cache3 8000 20 256
>> cache_dir aufs /sdb1/cache0 8000 20 256
>> cache_dir aufs /sdb1/cache1 8000 20 256
>> cache_dir aufs /sdb1/cache2 8000 20 256
>> cache_dir aufs /sdb1/cache3 8000 20 256
>>
>> maximum_object_size 18432 KB
>>
>> cache_swap_low 95
>> cache_swap_high 97
>>
>> access_log /usr/local/squid/var/logs/access.log squid
>> cache_store_log none
>> logfile_rotate 7
>>
>>
>> 2 36GB scsi disks are mounted on /sda1 and /sdb1 as
>>
>> /dev/sda1 on /sda1 type reiserfs (rw,noatime,notail)
>> /dev/sdb1 on /sdb1 type reiserfs (rw,noatime,notail)
>>
>> I can see lots of entries in cache.log that looks like:
>>
>> 2008/01/18 09:24:18| storeLocateVary: Not our vary marker object,
>> 1B384B5F6E91A0FE402D4DA6F9B067B0 =
>> 'http://x.tagstat.com/js/announcements.js',
>> 'accept-encoding="gzip,%20deflate"'/'gzip, deflate'
>>
>> Any suggestions please..
>>
>> Thanks
>> Manoj
>> --
>>
>
>
>
>

-- 
Received on Sat Jan 19 2008 - 23:13:06 MST

This archive was generated by hypermail pre-2.1.9 : Fri Feb 01 2008 - 12:00:05 MST