RE: FW: Request for CVS Branches.

Hello Henrick /Adrian

The Visolve Squid Team thanks you for giving them the developer access. We
will proceed from here and submit our developments as per the howtos. We
will contact you if we come across questions.

With thanks and best regards
Squid Team
www.visolve.com

-----Original Message-----
From: Henrik Nordstrom [mailto:hno@marasystems.com]
Sent: Saturday, May 11, 2002 3:12 AM
To: Squid Developers @ Visolve.com; adrian@creative.net.au
Cc: squid-dev@squid-cache.org
Subject: Re: FW: Request for CVS Branches.

Great!

You have hereby been given developer access to devel.squid-cache.org.
Please take some time to read the documentation on how to proceed (if
not, you mainly risk loosing your own edits..)

Required tools and howtos: http://devel.squid-cache.org/CVS.html
Some basic rules: http://devel.squid-cache.org/rules.html

If you have any questions, please ask.

Regards
Henrik

On Friday 10 May 2002 11:08, squid-dev@visolve.com wrote:
> Hello Henrick / Adrian
>
> We failed to mention in our previous mail given below that we also
> have a sourceforge account with account name "visolve"
>
> Expecting to hear from you on the CVS branches.
>
> Regards
> Squid team.
> www.visolve.com
>
> -----Original Message-----
> From: Squid Developers @ Visolve.com [mailto:squid-dev@visolve.com]
> Sent: Friday, May 10, 2002 1:06 PM
> To: adrian@creative.net.au; hno@marasystems.com
> Cc: squid-dev@squid-cache.org
> Subject: Request for CVS Branches.
>
>
> Hello Henrik / Adrian,
>
> This is regarding some of the development work in Squid undertaken
> by ViSolve team. We have partially completed the following two
> implementations.
>
> 1. RT Signal implementation.
> The objective of using RealTime signals to implement network I/O is
> to reduce CPU usage dramatically, thereby improving performance.
> The select/poll model must scan through a huge array of file
> descriptors to see which ones are ready for reading or writing.
> This scanning takes up lots of CPU time. With RealTime signals,
> read/write events for a socket are placed on a queue. To retrieve
> events, instead of polling, one just retrieves the first item in
> the queue.
>
> We have tested squid-2.4.STABLE3 and squid-2.4.STABLE3 with RT
> signal patch by disabling the disk. The results show very
> sigificant improvement in the performance. These two tests are
> taken with polymix-3 and hardware configurations are
> Processor : Pentium-3 933MHz
> RAM : 512MB
> NIC : 100Mbps card
>
> Our RT signal squid went upto 500 req/sec with CPU usage is below
> 80%, where as squid-2.4.STABLE3 reaches 100% CPU usage in 150
> req/sec itself. When we test our RT signal code by enabling the
> disk It is satisfying upto 230req/sec in the above configuration.
> During the test the CPU usage is less than 40% and memory usage for
> squid is 105MB, overall response time is 1.3 sec, miss response
> time is 2.5 sec and hit response time is .26 sec. This is 55% more
> than the normal squid.
>
> We have implemented this in Linux OS, kernel 2.4.13 by adding a
> patch for RT Signal. We have also identified other challenges which
> we think could be overcome with your help.
>
> 2. WCCP 2.0 compliance.
> Reference: Open source code developed by Joe Cooper in Squid 2.2
> for Wccp 2.0
> compliance. We have worked on the above said reference and have
> changed Squid 2.4 and
> Squid 2.5 for WCCP 2.0 Compliance.
>
> We request for 2 separate CVS branches for these codes or request
> suggestions from you on the
> procedure to submit the code and to share our futher developments
> in these two branches
> with Squid developers who would be interested to further this
> effort.
>
> Best Regards,
> squid team
> www.visolve.com
Received on Sun May 12 2002 - 21:52:17 MDT