[Bioclusters] Any issues porting applications to OS X?

[Tim Cutts]

On 6 Mar 2004, at 00:39, Farul Mohd. Ghazali wrote:

>
> Philip MacMenamin wrote:
>
>> If you take say a 16 node Altix 350 (I think thats the name of their
>> new low end machine), and compare it to a comparable 16 node cluster
>> using low latency interconnects, ie Myrinet, the Altix works out
>> cheaper than the Myranet cluster. (Based on the numbers that we got)
>> And it can run Linux. And is generally nicer.
>
> Is this a single system image machine or a cluster in the traditional
> sense? The MIPS based SGI Origins used some custom interconnects to
> create a nice NUMA machine across different nodes.

Yes, the Altix is single system image NUMA.  The 350 goes up to 16 
CPUs, the 3700 goes up to 128 CPUs in a SSI (although you can buy 
larger systems, they then have to be partitioned into 128 CPU images)

> If it's a single system, management and CPU/memory utilization would be
> significantly easier compared to the cluster but how well does Linux
> handle a NUMA environment?

A regular kernel doesn't.  As I understand it, SGI basically replaced 
the standard Linux scheduler with a NUMA-aware one, basically taken 
from IRIX.

The Altix is basically an Origin but with Itanium2 rather than MIPS 
CPUs.  In fact I think the Altix 3700 and Origin 3000 are even hardware 
compatible, to the extent that the router and I/O bricks are the same, 
it's only the memory and CPU bricks which differ.

I'm trying to get my hands on a 350 to test at the moment.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group
Wellcome Trust Sanger Institute
Hinxton, Cambridge, CB10 1SA, UK