Re: [squid-users] Request for your recommendation for ISP setup

From: Khemara Lyn <lin.kh_at_wicam.com.kh>
Date: Fri, 04 Jun 2010 22:32:41 +0700

On 06/04/2010 08:13 PM, Luis Daniel Lucio Quiroz wrote:
> Le vendredi 4 juin 2010 00:09:19, Khemara Lyn a écrit :
>
>> Dear All,
>>
>> First, please bear with me for the lengthy message. I'm really in need
>> of help from your expertise regarding a good, robust, high-performance
>> forward-proxy Squid setup for ISP customers.
>>
>> I am running an ISP with around 500 customers. I've been using a single
>> Squid machine to do forward proxy for the customers to cache the Web
>> contents and thus save some costly bandwidth.
>>
>> The single Squid machine has the following hardware specs roughly:
>>
>> - RAM: 16GB
>> - CPU: 2 of 3 GHz Intel XEON CPUs
>> - Hard drive: 4 x 300GB SCSI drives
>>
>> I use Squid-2.7STABLE9 on Fedora 12.
>>
>> Right now, I allow only half of the customers (around 250 users) to use
>> this forward proxy machine and I notice that, the 16GB memory is used up
>> easily in 3 hours after Squid's startup.
>>
>> I would like to know how can tweak that box for better performance than
>> it has now.
>> Or is it reaching the limit already?
>>
>> Please find in the attached files for the Squid configuration, and cache
>> info& utilization.
>>
>> I am also thinking of running 2 Squid machines as cache peers: one being
>> a child and the other a parent. For that setup, I would like to have the
>> child peer to do caching for local customers and redirect any outside
>> (Internet) destinations to the parent peer, which will not cache anything.
>>
>> May I have your inputs on this setup: is it correct and does it follow
>> the best practice?
>> If it does, may I have some guidances/pointers on this from those who
>> had set up similar scenario before?
>>
>> Hope for your kind advice.
>>
>> Many thanks& best regards,
>> Khem
>>
> Hi there
>
> well
> i would move these
> cache_mem 6 MB
> maximum_object_size_in_memory 32 KB
> minimum_object_size 512 bytes
> maximum_object_size 5 GB
> to these
>
> cache_mem 12 MB -> if you are only runng in that machine one squid
> maximum_object_size_in_memory 64 KB -> 69% of objects are less than 64kb
> minimum_object_size 0 -> let all
> maximum_object_size 135128 kB -> >0,01% of objets are more than 135mb
>
>
>
Thank you for your useful comments. I will try to change as per your
suggestion and observe it for a while.
> refresh_pattern . 0 20% 4320
> should e recalculated but i tis no trivial
>
> and i think you are missing "cache" lines to let the videos/mp3 let in as you
> configuration states.
>
Sorry, I don't quite understand your comment here. I don't know how I
can add the "cache" lines to let "video/mp3".

I have this my squid.conf:

refresh_pattern -i
\.(mp2|mp3|mid|midi|mp[234]|wav|ram|ra|rm|au|3gp|m4r|m4a)(\?.*|$)
5259487 999% 5259487 override-expire ignore-reload reload-into-ims
ignore-no-cache ignore-private

  What else should i add?

> Caché dir, dont give more than 250 GB of space, and 32 256 values should be
> recalculated to let tree navigation faster as possible
>
> however your statics
> Request Hit Ratios: 5min: 29.6%, 60min: 29.1%
> Byte Hit Ratios: 5min: 22.6%, 60min: 24.2%
>
> are far enoguht so dont expect more performance you are about to reach the 30%
> that statistically maximum you may reach (in general speaking).
>
>
I could see the hit ratios and overall bandwidth saving was good but not
happy with the way it eats up memory so fast and become slow after 3hours.

> LD
>
>
>
>

How about the cache peer (child and parent setup)? Anyone has comments
or some pointers to a good document about it?

Thanks again,
Khem
Received on Fri Jun 04 2010 - 15:32:59 MDT

This archive was generated by hypermail 2.2.0 : Fri Jun 04 2010 - 12:00:04 MDT