[Top] [All Lists]

Re: productionserver

To: "Linux XFS Mailing List" <linux-xfs@xxxxxxxxxxx>
Subject: Re: productionserver
From: "Steve Wolfe" <nw@xxxxxxxxx>
Date: Thu, 23 Aug 2001 16:48:01 -0600
References: <Pine.BSI.4.10.10108232336290.21190-100000@xxxxxxxxxxxxx>
Sender: owner-linux-xfs@xxxxxxxxxxx
> We are rather diskbound with our server rather then processor.
> The new machine will feature a dual 1Ghz PIII with at least 1GB of ram
> the raid will be a raid 10 totaling 6 disks with 2 raid0 sets which are
> mirrored. The disks will be 73GB 10K RPM disks and each set of 3 will be
> on it's own channel on the raid controller.

  You might want to consider the dual Athlons.  My gut feeling is that for
database work, they will smoke P3's.  In a dual P3, you're sharing a 133
MHz bus between both processors, giving each CPU an effective 66 MHz bus
if they're both working at full tilt.  With the Athlons, each CPU has an
independant 266 MHz bus, and even sharing the bandwidth to the memory,
they still have an effective 133 MHz each, due to the DDR RAM.  So, for
working with large tables, where you have to move very large amounts of
data to/from RAM in a short amount of time, having twice the bandwidth
will probably make a very large difference.

> In case you want to order one, it's a Dell PE 2500 with rackmount kit an
> 6 73GB disks, UPS, and 3 year support. This brings the price in the
> Netherlands to about 18.000 Euro (divide by 1.1 for US $).

  I've used Dells before, and while they're decent machines, the machines
that I build myself are less expensive, just as reliable, and faster.  For
less money than that, we put together a quad Xeon, with SCA backplanes,
hardware RAID, the works, and that was over a year ago.  Of course, I
realize that being in Europe probably increases your cost, and decreases
your selection.

 (Incidentally, I'm thinking of testing a dual Athlon to replace the quad
Xeon.  The MHz rating would be similar (2.4 GHz to 2.8 GHz), but under
heavy load, each Xeon has the equivalent of a 25 MHz bus, so even when the
machine is maxed out, the bus seems to be the largest bottleneck.  I don't
think I've ever seen the CPU's hit more than about 80% utilization,
they're sitting around waiting for data.


<Prev in Thread] Current Thread [Next in Thread>