Re: [squid-users] Multiple Accelerators.

As Henrik said, the ufs on tmpfs method will give you a directory tree
you can push out on shutdown and pull from on bootup. (This would be
outside of squid, but just a 'cp -a' in your start/stop script will
suffice.)

If you don't mind clearing your cache on restart, null fs will be
faster, since there's no syscall or filesystem overhead.

        -- Brian

On Monday 03 March 2003 07:47 pm, SSCR Internet Admin wrote:
> In addition to his question, brian.. Is it possible before a system
> shutdown squid will save all cache objects from RAM to disks? And
> load all cache object from object to RAM? coz you have /dev/null as
> the cache directory...
>
> -----Original Message-----
> From: Brian [mailto:hiryuu@envisiongames.net]
> Sent: Sunday, March 02, 2003 11:55 PM
> To: Allan; squid-users@squid-cache.org
> Subject: Re: [squid-users] Multiple Accelerators.
>
> On Saturday 01 March 2003 05:55 pm, Allan wrote:
> > Hello,
> >
> > we are currently using a single server (squid 2.4-STABLE7, Linux
> > RedHat 7.3, 1.4 GHz Pentium III, 2 Gb Ram, 4096 FileDescriptors) as
> > reverse-proxy for a small site (approx 2,5K objects, about 150 Mb).
>
> To extract the most out of this box, try
> * Squid ufs over Linux tmpfs or
> the null fs and large cache_mem & maximum_object_size_in_memory
> * If this uses an Intel NIC, replace it with a 3com
> * A kernel update -- the 2.4 jam patches are very nice.
>
> > Due to growing load - 400 hits/sec, we are experincing loads about
> > 30% user and 65% system-time
> >
> > Is this alarming? Should we consider bying "web"-switches and
> > adding furter servers?
>
> The system time seems a bit high for a squid server that should be
> 100% in memory. The NIC and the filesystem are the likely culprits.
> At 400 req/sec, I would consider a trio of servers just for smoothing
> over uprades or other downtime. Obviously the budget is not always
> so flexible.
>
> > As the site is quite small, would it be a good idea to build a
> > "mini"-linux booting of floppy/CD with no harddrives/filesystems?
>
> Not really necessary -- with that much RAM, yous shouldn't be
> touching the hard drive, anyway. I would go with null storage or
> tmpfs for the squid cache, though.
>
> -- Brian
Received on Mon Mar 03 2003 - 11:55:05 MST