Top of the line Shared Memory

[james_lindsay]

Hello,
I just won a grant to purchase a computer system to support the University of Connecticut's 454, and Solid (comming soon) sequencing machines. Currently we are involved in sequencing several genomes, and are attempting to perform de novo assembly.
I was wondering what system others have purchased to do large mammalian de novo assembly, under 100K.

Our current 128GB Dell runs out of memory while running velvet (on a combination of Sanger, 454 and Solid data).

Currently I think the dell R900 has 256GB memory.
Sun has a machine with 512GB
SGI Altix 3700BX 64 1.5Ghz with 512GB RAM.
Has anyone had any experience with scaleMP to build a virtual high memory system?

Thanks,
James

[fultheim]

454 performance with ScaleMP's vSMP Foundation

Hello James,

Please check our web-site at: www.scalemp.com/performance - we included some 454 performance results for your convenience.

--shai

[james_lindsay]

Those graphs don't really answer my question. It would be good to see a direct comparison between a real shared memory system. Intuitively one would think the performance would go as follows:
1- Sun system with 512GB
2- SGI Altix with 512GB
3- scaleMP with 512GB

The sun system has the memory most tightly integrated and closest to the processors. The Alitx has the memory distributed through the SGI interconnects.
Finally scaleMP has the memory distributed through infiniband connections.
I think this is confirmed at http://www.cs.virginia.edu/stream/top20/Bandwidth.html

I would love to go the scaleMP route but I can't find anyone who has independently benchmarked bioinformatics applications. I would love to chat more with you or a technical consultant from your company but no one has answered my email requests for more information.

Also the newbler comparison seems to be showing to performance of RAMdisks vs harddisks, not comparing run time of a scaleMP system and a traditional shared memory system.

You can contact me at [email protected]

[fultheim]

James,

There are performance comparison for large memory application such as NASTRAN on the performance link I sent before. Indeed, one of the advantages of very large memory systems is the ability to use RAM rather than just plain (and slow) drives. This should be considered as a plus.

I'll try to connect you with other bio-informatics users.

[gardner]

Raises Hand

Hi James,

We have a ScaleMP system down here at UF and are using it for next gen sequencing projects. Shoot me an email ([email protected]) if you would like some more information.

- Aaron

--
Aaron Gardner
IT Expert - Computational Infrastructure
Interdisciplinary Center for Biotechnology Research
University of Florida P.O. Box 103622
152 CGRC
Gainesville, FL 32610

Email: [email protected]
Web: www.biotech.ufl.edu
--