Re: [squid-users] Re: question about large rock

From: Alex Rousskov <rousskov_at_measurement-factory.com>
Date: Thu, 13 Feb 2014 12:43:07 -0700

On 02/13/2014 03:58 AM, Dr.x wrote:
> hi all ,
> im asking about memeory caching
> now we know that we have restirction with smp rock to 32 k , whiuch result
> a poor bw utilization

It is really not clear what you mean by "poor bw utilization" and how it
is related to 32KB object size limits in small rock. I recommend that
you use a different terminology or explain the connection between the
two concepts so that it is clear what you are asking about. Perhaps you
mean poor byte hit ratio?

> now
>
> what about memory object size ?
> is is also 332 k ?

Shared memory cache prior to Large Rock support has the 32KB limit.
Shared memory cache after Large Rock support does not have that limit.
Non-shared memory cache does not have that limit.

> i mean if i get a server and depened on ram and have 128G ram , and dont
> want to use hardsisks
>
> does in this case we still have 32 k object size max in ram ??

Depends on the memory cache type. See above and shared_memory_cache in
squid.conf.documented.

> what is the best implementation for smp with rock with aufs so that we have
> max bw utilization and max hit ratio ?

Rock and aufs are two different cache store modules. You cannot use
"rock with aufs". Rock was designed for SMP use. The current aufs store
implementation is not really compatible with SMP. The best store to use
depends on many factors. For example, you can use Large Rock store to
avoid the 32KB limit but you will have to run Squid trunk, with all the
caveats that implies.

HTH,

Alex.
Received on Thu Feb 13 2014 - 19:43:13 MST

This archive was generated by hypermail 2.2.0 : Fri Feb 14 2014 - 12:00:04 MST