Re: [squid-users] Squid on DualxQuad Core 8GB Rams - Optimization - Performance - Large Scale - IP Spoofing

On 17/10/2007, at 10:50 AM, Muhammad Tayseer Alquoatli wrote:

> On 10/17/07, Andrew Miehs <andrew@2sheds.de> wrote:
>> On 17/10/2007, at 10:03 AM, Muhammad Tayseer Alquoatli wrote:
>>> what about AUFS, is in it a way to utilize the other cpus ? i'm
>>> running an SMP machine with aufs and when i issue mpstat -P ALL i
>>> see that all cpus are used (not equally but nearly equally) and
>>> there is no other application running on the box but squid?
>>
>>
>> Let me guess - if you add the usage on all your CPUs together you do
>> not go over, or not far over '1' ?
>>
>
> Hi Andrew
>
> yes, probably, i don't remember i've got more than 1, but the overall
> load is little (0.25-0.3), do you have an explanation ?
> Thanks

If I understand aufs correctly, it is that squid uses threads rather
than
a queue to store the 'to be written requests'.

   "aufs" uses the same storage format as "ufs", utilizing
   POSIX-threads to avoid blocking the main Squid process on
   disk-I/O. This was formerly known in Squid as async-io.

If you only have '1' set of disks you will probably find that this will
not give you any improvement.

Looking at your load I assume you have 4 processors (probably 2x Dual
Core)
or 2x Hyperthreading CPUs - correct?

The best way that I know of you increase your squid performance is
add more
disks. I haven't tested it (as my squids run RAM (16g) only) - but
you will probably be better off with 4 separate filesystems on 4 disks,
rather than one raid 10 on the same for disks -
ie: (The sizing and filesystem is just copied from example from the web)

cache_dir ufs /cache/disk1 200 16 256
cache_dir ufs /cache/disk2 200 16 256
cache_dir ufs /cache/disk3 200 16 256
cache_dir ufs /cache/disk4 200 16 256

Cheers

Andrew
Received on Wed Oct 17 2007 - 03:12:28 MDT