Re: [squid-users] squid in ISP

The tool to use is polygraph, but the workloads may need updating to
match currently seen traffic patterns.

The tools are there - whats needed is equipment, time and manpower.
The project is lacking all of the above.

Adrian

2008/7/18 Richard Hubbell <richard_hubbe11_at_yahoo.com>:
> --- On Thu, 7/17/08, Amos Jeffries <squid3_at_treenet.co.nz> wrote:
>
>> From: Amos Jeffries <squid3_at_treenet.co.nz>
>> Subject: Re: [squid-users] squid in ISP
>> To: richard_hubbe11_at_yahoo.com
>> Cc: lsk_at_rocketmail.com, "Rhino" <rhino_at_machlink.com>, squid-users_at_squid-cache.org
>> Date: Thursday, July 17, 2008, 6:26 AM
>> Richard Hubbell wrote:
>> >
>> >
>> > --- On Fri, 7/11/08, Rhino <rhino_at_machlink.com>
>> wrote:
>> >
>> >> From: Rhino <rhino_at_machlink.com>
>> >> Subject: Re: [squid-users] squid in ISP
>> >> To: lsk_at_rocketmail.com
>> >> Cc: squid-users_at_squid-cache.org
>> >> Date: Friday, July 11, 2008, 6:56 AM
>> >> Siu-kin Lam wrote:
>> >>> Dear all
>> >>>
>> >>> Any experience using squid as caching in ISP
>> >> environment ?
>> >>>
>> >>> thanks
>> >>> SK
>> >>>
>> >>>
>> >>>
>> >>>
>> >>>
>> >> I'm sure there's much larger ISPs out
>> there and
>> >> been using it much longer;
>> >> just passing along our info.
>> >> We're a small ISP serving around 10k
>> dialup,dsl,cable
>> >> modem and MAN subs
>> >> via a dual-homed to different ISP BGP WAN.
>> >> We loaded squid on a quad core linux box with
>> around 1.2Tb
>> >> disk
>> >> capacity and 32Gb RAM, using a Cisco 4948 switch
>> and WCCP2
>> >> to transparently redirect to Squid.
>> >> There were some major hurdles along the way
>> >> mostly getting the 4948 to pass the L2 WCCP
>> traffic -
>> >> 2 IOS bugs and a year in the process) but once
>> that worked
>> >> and we got our IPTABLES set up properly,
>> transparent
>> >> redirection
>> >> has been working quite well.
>> >> Some tweaks needed to our Squid config, but with
>> the help
>> >> of this list
>> >> - particularly Henrik and Amos' posts - at
>> this point
>> >> we're very
>> >> encouraged by the performance and bandwidth
>> savings
>> >> we're seeing on the
>> >> system which has only been truly active for around
>> 3 weeks
>> >> now.
>> >> Again, we're a pretty small shop - so when our
>> old
>> >> NetApp Netcache
>> >> was no longer able to adequately handle the load,
>> we needed
>> >> an
>> >> effective, minimal-cost solution which this is
>> >> demonstrating to be.
>> >> Hope that helps.
>> >> -Ryan
>> >
>> >
>> > Thanks for sharing this. We're doing about 75
>> requests/sec on a quad-core Xeon with 16GB. Still trying
>> out some different configs.
>> > I have cache_mem set to 2GB and it's working well
>> so far.
>> >
>> > It's not even worked up a sweat and has plenty of
>> room for more work.
>>
>> I'll bet it isn't.
>> 75 is not even close to half what squid was doing in Y2K.
>> :)
>>
>> If you want to stress it we'd be glad of the results.
>
> I can do this. Is there a set of tests that people would like to see?
> I think we should concoct a test plan, albeit a very basic one.
>
> Is there a squidbench? squib for short? Maybe it could start by either fetching a digest or using one as input? Or just use a list of urls from an access.log?
>
> If we don't have a level SUT (system-under-test playing field) we can at least have a level load-generator. The fewer variables the better.
>
>>
>> Amos
>> --
>> Please use Squid 2.7.STABLE3 or 3.0.STABLE7
>
> Are there redhat packages for 2.7?
>
>
>
>
>
>
Received on Fri Jul 18 2008 - 06:32:41 MDT