[Bioclusters] Re: Recommendations for building a cluster

Chris Dagdigian dag at sonsorol.org
Sat Oct 20 09:17:07 EDT 2007


Hi Ahmed,

I have personal experience with Mac clusters and I just finished  
building a moderately sized cluster built from the Sun Fire X2100 M2  
systems. I've not used Dell rackmounts in a very long time but not  
for any negative reasons.

My personal recommendation is to pass on Apple for now -- the Intel  
Xserves are a few CPU generations behind now and it seems the new  
focus of Apple as a consumer products company means that they are not  
really focusing on "enterprise" products meant for datacenters any  
more. The only alternative I'd say with Apple if you are thinking  
about a smaller cluster would be to build one from their workstation/ 
desktop line which does track the modern Intel releases pretty closely.

The Sun Fire X2100 M2 machines were fantastic, the remote management  
ILOM processor is great and allowed us to configure and build the  
cluster remotely using nothing but VPN and SSH clients.

If you go the Dell route, consider paying extra for the Dell version  
of the remote management card (I think they call it a 'DRAC' ? )

Another hint -- wait until mid-November or later to really nail down  
your purchase. The SuperComputing 2007 show is coming up shortly and  
there are going to be some interesting things and new products  
released at that show. You'll also start to see the quad-core/quad- 
CPU motherboards show up in vendor products -- 16 cores in a single  
chassis may be interesting for the type of work you are doing.

-Chris




On Oct 19, 2007, at 6:23 PM, Ahmed Moustafa wrote:

> Thank you so much Chris!
>
> By "genome-wide", it is more like performing the same task (e.g.  
> blast, clustalw and phyml) for every single gene in a genome. But  
> because a single task could include +200 sequences to align or to  
> build a phylogeny, it takes a significant amount of time and memory  
> to process a single gene. So by using, for example, mpiblast and  
> raxmlmpi, it could be possible to distribute these tasks over a  
> cluster and finish an analysis in a reasonable time.
>
> I have googled and found these options Dell PowerEdge SC1435  
> (http://www.dell.com/content/products/productdetails.aspx/ 
> pedge_sc1435), Sun Fire X2100 M2 (http://www.sun.com/servers/entry/ 
> x2100/) and Mac Cluster (http://www.apple.com/science/solutions/ 
> workgroupcluster.html), do you have an experience with any of these  
> options?
>
> Thanks again!
>
> Ahmed



More information about the Bioclusters mailing list