Re: Is squid unable to handle the load?

Morten Guldager Jensen wrote:
>
> On Wed, 27 May 1998, Markus Storm wrote:
>
> > On Tue, 26 May 1998, Mark Dabrowski wrote:
> >
> > > As you can see when the disk operations / second go above 100 (12th
> > > column), the CPU idle time goes 100% (last column), which basically means
> > > squid freezes...
> > >
> > > Correct me if I'm wrong, but that means our SCSI controller (Adaptec 2940
> > > Ultra) / disk is maxing out at 100 operations / second...
> >
> > Not sure about that one, but decent controllers do 2000 ops or so ....
> > Given an average seek time of 10 ms per access (remember it's mostly
> > non-clustered, single random accesses), you're maxing out the disk.
> >
> > Try adding more disks.
>
> Will it help? I don't think so.
>
> As far as I know, Squid does sequential disk IO.
>
> So when Squid decides to read a file, everything else has to wait. Then
> you are waisting the additional disk IO bandwidt comming from the
> multiple disk drives.

Yes and no.
First, don't run anything else on the disks you're putting the cache data on.
Second, yes, squid 1.1 does sequential I/O. Squid 1.2 does it asynchronously.
It's not 100% stable yet but one might nevertheless consider running it (we
do).
Third, in case 1.2 is not an option, striping spreads the load across several
disks which greatly decreases the average access time, thus allowing more
(sequential) accesses per second.

Markus
Received on Wed May 27 1998 - 05:09:25 MDT