Hi
Frontier Squid makes heavy use of the IMS (if modified since) feature, which doesn't work as advertised in the recent 3.x versions of squid-cache. There is a bug report filled. Until this is fixed we are stuck with these packages which are indeed ugly but work. I suspect that several sites are also using old versions of the frontier-squid package and the newer versions have many fixes. The frontier squid package works fine on SL6.
While we are on the subject of frontier squids. ATLAS have recently been validating "overlay Monte Carlo". These are MC simulated signal events overlaid with background from real (minimum bias) events. These types of jobs use on average around ~200MB of conditions data, compared to a normal analysis job needing somewhere between 10 - 70 MB. Obviously these types of job won't be run all the time but usage of Frontier squids is certainly growing. Large sites (currently CERN/BNL/Fermilab) are looking to deploy frontier squids with 10GBit/s links.
Alastair
On 28 Sep 2012, at 17:00, Ewan MacMahon wrote:
>> -----Original Message-----
>> From: Testbed Support for GridPP member institutes [mailto:TB-
>> [log in to unmask]] On Behalf Of james Adams
>>
>> There are so many things wrong with the frontier squid package, I would
>> very strongly recommend only using it for frontier squids and using the
>> normal squid package for CVMFS and everything else.
>>
> Could you be a bit more specific? Having a common squid is quite nice
> for the smaller sites.
>
> Ewan
|