On 08.05.2012 07:07, Pieter De Wit wrote:
On 8/05/2012 00:56, Mário Sérgio Candian wrote:
Hi Peter.
Thanks for the answer.
I'd like to run FreeBSD in this server. I don't tried any config.
I'll buy
this server but I need to know if this server supports the amount of
users
that I have.
Yes, this will be a cache for my ISP. The server has gigabit network
card. I
have a link of 400Mbps.
This server can handle this amount of users? 15000 users? About the
configuration squid, what do u recommend?
Regards,
MSC
Hi,
I would buy more cache drives (assuming you going to be using the SSD
for this - change the 300gig drives to 146gig's to save some money,
heck, you can even change it to a mirror set of 72gig, it's only the
OS). I won't mess with the default settings since Squid is pretty
tuned as it is. I have ran 6000 connections thru boxes smaller than
that, way smaller. The only time I would change those settings is
when
you need to force more caching out of it.
I would also research the disk to memory caching formula on the Squid
wiki (work out how much memory X gig on disk cache needs). More that
this I can't offer without more input from you, otherwise I might as
well deploy the box.
Cheers,
Pieter
The two main hardware bottlenecks for high performance Squid is CPU GHz
and disk RPM numbers. The higher those numbers the better. Squid is
event-driven software instead of threaded, so multiple 1-2 GHz cores is
worse than a single 3+ GHz core. Higher disk RPM of course speeds up the
delivery of disk cached items, so 15K RPM disks better than larger 10K
RPM even when smaller (there is a 2^24 or so object limit per disk which
limits utility of big disks anyway).
With an 8-core machine you will likely be looking at a multiple-process
setup for Squid to achieve full 400Mbps (3-5 squid processes?). 3.2
series can do this natively with workers, older Squid need some slightly
complex config. Details are in the wiki.squid-cache.org documentation
under SMP and multi process support.
Amos
