Re: [SQU] high throughput?

Florin Andrei wrote:

> >From: dualcat@163.net
> >
> >Dear friends,
> > I like squid.From the benchmark results,it seems that the max peak
> >throughput of squid is near 140 req/sec with x86 systems.But,in my
> >application,I need a cache system which peak throughput reach 1000
> >req/sec.Can I do that with squid?
> >with modify the code or change the OS or change the hareware?
>
> Might be possible if using some real powerful hardware, like an SGI box
> with many processors (SGI's are really good at throwing a lot of bytes
> through any I/O channel) or maybe a multiCPU Alpha.
> But i have a better idea: just install 8...10 x86 boxes, all of them
> running Squid, and put a Linux Virtual Server in front of them all. I tried
> this, it works outstandingly.
> If you need more power, just add some more x86 systems. LVS will virtualize
> the service.
> I suggest to use the Direct Routing implementation of the Virtual Server,
> because it offers the highest performance.
>
> http://www.linuxvirtualserver.org/
>

or you can use freebsd memory file system (MFS) for your cache partition instead of
using disk. but of course you need a lot of ram here aside from gigabit ethernet and a
higher processor for a single box without modifying the code of squid for a higher
throughput. and also, you already know the consequences for having an MFS without using
a good backup management when power failure comes. but try to weight the price over
performance of multiple squid boxes in front of an L4 switch versus to a single box as
your http traffic grows.

fooler.

--
To unsubscribe, see http://www.squid-cache.org/mailing-lists.html