Re: [squid-users] cache size and structure

From: Riccardo Castellani <ric.castellani_at_alice.it>
Date: Fri, 26 Jun 2009 22:27:47 +0200

>>> My cache traffic volume (I/O) is about 2 Mbps a week with peaks of 3
>>> Mbps.
>
> ehm, 2 megabits per socond "a week"?

excuse me, 1 megabit per second for a week, so :
1 Mb *60*60*24*7=604.800 / 8 = 75.600 MB about 76 GB traffic for week.
I'm refering both inbound and outbound traffic (1 Mbps in , 1 Mbps out).

>If users acess your squid directly (not only through child proxies), there
>may be the need for cache_mem.

some users access by child squids

>I have "48 256" for 20GiB cache. If you take the number of files in the
>cache directory, divide by 256 (l2 dirs) and 256 (max files in l2 dir), you
>should get the approximate need of l1 dirs. The average object size is
>aropund 13KiB, which means, you should have one L1 cache_Dir per ~800MiB of
>cache size. Splitting small files to COSS (using min-size option for *ufs
>and max-size for COSS) will make that even smaller number, since only big
>iles will be placed to *ufs directory hierarchy.

I'm sorry but I didn't understand.
How can I enable COSS ?

----- Original Message -----
From: "Matus UHLAR - fantomas" <uhlar_at_fantomas.sk>
To: <squid-users_at_squid-cache.org>
Sent: Thursday, June 25, 2009 11:02 AM
Subject: Re: [squid-users] cache size and structure

>> Riccardo Castellani wrote:
>>> I'm preparing new squid machine and I'm defining cache size.
>>> Old squid had 2 entries into "cache_dir directive" :
>>>
>>> cache_dir ufs /usr/local/cache/1 3500 128 256
>>> cache_dir ufs /usr/local/cache/2 2500 128 256
>
> On 23.06.09 13:07, Chris Robertson wrote:
>> I'd strongly suggest using "aufs" instead of "ufs".
>
> seconded.
>
>>> My cache traffic volume (I/O) is about 2 Mbps a week with peaks of 3
>>> Mbps.
>
> ehm, 2 megabits per socond "a week"?
>
>>> This squid cache is the parent of 2 other squid machines and it gives
>>> answers to about 1000 users.
>>>
>>> 1- I read you suggest 1 cache_dir to same partition, why I can't use 2
>>> folder in tha same partition ?!
>
>> You can, but why would you want to? The suggestion is one cache_dir per
>> spindle to spread the IO load. Putting multiple partitions on one
>> spindle makes about the same sense as multiple cache_dirs in the same
>> partition. Access to all of them will be contending for the limited IO
>> resources available.
>
> That means, different cache_dirs are mostly for using different disks.
> Or different storage schema, e.g. using COSS (which is perfect for small
> files) in addition to *ufs
>
>>> 2- What do you think my caches size ? 3500 and 2550 ?
>>
>> Depends on your memory load. A larger cache leads to storing more
>> objects, which requires more memory to track. The suggestion I recall
>> is "a week's worth of traffic". If you are seeing an average of 2Mbit/s
>> 24 hours a day, seven days a week, that would lead to a cache of around
>> 150GB.
>
> I'd say it depends on disk load also :) caching one-two weeks of HTTP
> traffic is not that good if disk becomes the leak. But switching to one
> aufs cache_dir (plus optional COSS) should give enough of speed benefits
> so
> the cache could be increased.
>
> If users acess your squid directly (not only through child proxies), there
> may be the need for cache_mem.
>
> See please squid FAQ for informacions about memory usage...
> http://wiki.squid-cache.org/SquidFaq/SquidMemory
>
>>> and its directory structure (128,256) ?!
>
>> This is entirely dependent on the filesystem you are using and the
>> number of objects you cache. The goal is to keep the number of files
>> per directory reasonable, because most filesystems are not optimized for
>> a "large" ratio (10s of thousands of files per directory).
>
> I have "48 256" for 20GiB cache. If you take the number of files in the
> cache directory, divide by 256 (l2 dirs) and 256 (max files in l2 dir),
> you
> should get the approximate need of l1 dirs. The average object size is
> aropund 13KiB, which means, you should have one L1 cache_Dir per ~800MiB
> of
> cache size. Splitting small files to COSS (using min-size option for *ufs
> and max-size for COSS) will make that even smaller number, since only big
> iles will be placed to *ufs directory hierarchy.
>
> --
> Matus UHLAR - fantomas, uhlar@fantomas.sk ; http://www.fantomas.sk/
> Warning: I wish NOT to receive e-mail advertising to this address.
> Varovanie: na tuto adresu chcem NEDOSTAVAT akukolvek reklamnu postu.
> Emacs is a complicated operating system without good text editor.
Received on Fri Jun 26 2009 - 20:28:07 MDT

This archive was generated by hypermail 2.2.0 : Sat Jun 27 2009 - 12:00:03 MDT