[rescue] intel vs. sun- for real

[Patrick Giagnocavo +1.717.201.3366]

On Sun, Jul 21, 2002 at 01:01:27AM -0400, s at avoidant.org wrote:
> Koyote wondered:
> 
> >  So, my girlfriend is being forced to give hard evidence why a
> >  single or dual cpu UltraSparc would be more advantageuous to
> >  purchase than an intel box. Performance, reliability, anything.
> 
> 
> OK, I've got a similar question. I'm wanting to replace 8 Intel boxes
> running Solaris with SPARC(s). Can someone recommend a suitable machine
> that won't cost upwards of $100,000?
> 
> The machines in question are each attached to eight 73G drives, and suck
> backups from the client servers on our network. They're doing this
> fairly constantly, day and night, and gzipping the data (keeping the
> load off the client boxes). The move the data off to tape once a week
> for (semi)permanent archiving.
>

I think the first thing you need to do is to take a sample Intel box
and a sample SPARC box and test how long it takes to gzip a typical
file.  

I believe that the I/O throughput is not a problem, but that the
bottleneck could well be in gzip. Without having an idea of gzip
performance you won't have any idea how it will work until you put it
in production, which would of course be a bad thing.  

You might want to try different gzip binaries compiled with different
options, since even a few optimizations could improve performance
substantially.

Once you have an idea of how the CPUs compare in the gzip task you can
figure out from there which SPARC machine is most suitable.  Maybe you
will only need 2 CPUs in a box to match the performance - so you could
specify a machine with 4 and have lots of room to grow.
 
> So we're talking major I/O, and high cpu load. Almost anything
> affordable will be on it's knees.

You might end up quite surprised. :-)

Cordially

Patrick Giagnocavo
patrick at zill.net