dual AMD clusters

[Art Edwards]

On Fri, Jun 22, 2001 at 12:04:38PM -0700, Martin Siegert wrote:
> On Thu, Jun 21, 2001 at 16:45:27 -0400, Dan Kirkpatrick wrote:
> > We're about to build our PoP (Pile of Pentiums) beowulf cluster #2 and are 
> > trying to build this one even better since our budget is 4x that of the 1st 
> > one...
> > There's a wide variety of needs, some small code, some large (200mb-1gb+).
> > Mostly threaded, some in the future may be parallel.
> > 
> > We've got a few questions for the list...
> <snip>
> > 3. Processor choice
> > What about Dual Athlon's?  Are they actually available and reliable?  We've 
> > gotten the feeling that they're fairly new, run hotter (more cooling/larger 
> > case needed), and more expensive, but for the calculations we're doing, it 
> > may be more power for the money in the end if they are actually available 
> > and reliable.
> 
> I have almost completed my test of a dual AMD cluster. The test cluster
> had two dual nodes. The master node has 5 NICs (2 onboard 3c980, 3 3c905B),
> the internal node has 4 NICs (2 onboard 3c980, 2 3c905B). 3 NICs are used
> in a channel-bonded configuration, one is used for NFS traffic. The
> remaining one on the master node is for the internet connection.
> 
> I am using a 2.4.5 kernel on a otherwise RH7.1 system.
> 
> I have encountered a few problems, all of which, but one have been solved:
> 1) Pay close attention to the memory chips that Tyan has approved. Other
>    chips may not work.
> 2) The latest BIOS upgrade solves problems with booting the nodes.
> 3) There is a bug in the 2.4.5 kernel (somewhere in the apic code) that
>    brings network connections to a grinding halt. Using the -ac versions
>    (e.g., 2.4.5-ac17) solves this problem.
> 4) lm_sensors fails to recognize the hardware monitoring chips on the
>    Tyan motherboard.
> 
> Otherwise the system has been rock solid: no crashes, very good performance.
> Furthermore, Tyan is going to release a no-SCSI version of the motherboard
> soon - this will make the dual AMD system very competitive - I finally
> made up my mind to go that way.

I am still considering whether to go with single or dual CPU motherboards in 
an upgrade to the current Beowulf. Memory ber CPU is very important for us.
Have you done benchmarks of single and dual CPU nodes? Are two really twice 
as fast as one? I will be running MPI-based parallel codes.

Art Edwards
> 
> If only somebody could show me how to patch lm_sensors to detect the
> hardware monitoring chips on the Tyan motherboard:
> 1) I was able to insert the i2c-amd756.o module (after changing the
>    PCI_DEVICE_ID to 7413).
> 2) sensors-detect now shows a Winbond W83782D chip, but "modprobe -k w83781d"
>    brings the box to a full stop (only a hard reset helps). This obviously
>    needs some work - does anybody know more about this?
> 
> Cheers,
> Martin
> 
> ========================================================================
> Martin Siegert
> Academic Computing Services                        phone: (604) 291-4691
> Simon Fraser University                            fax:   (604) 291-4242
> Burnaby, British Columbia                          email: siegert at sfu.ca
> Canada  V5A 1S6
> ========================================================================
> 
> _______________________________________________
> Beowulf mailing list, Beowulf at beowulf.org
> To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf

-- 
Arthur H. Edwards
712 Valencia Dr. NE
Abq. NM 87108

(505) 256-0834