[Bioclusters] Need some advice on a cluster for EST/cDNA assembly, clustering

[Tim Cutts]

On 28 Feb 2005, at 4:44 pm, Dr D.S-M. Huen wrote:

> On Feb 28 2005, Matthew Hudson wrote:
>
>> Another key consideration is: How much sysadmin experience / effort 
>> does this life science group have? Maintaining and updating a Linux 
>> cluster built from scratch can be a challenge for a group of biology 
>> PhDs with no full-time sysadmin, not to mention the cost of power 
>> consumtion, cooling etc.. Sun is a little better than building your 
>> own Linux system, Apple really try to make this easier and might be a 
>> better solution, but their cluster software is expensive. If it's 
>> just an EST project you might recommend they get one or two good size 
>> SMP machines (maybe the new quad Opterons from Sun, with at least 8GB 
>> RAM) and forget about the cluster.
>
> I must admit I am considering whether I should aim to use something 
> like Sun's Compute Grid service in future ($1/ CPU-hr). The overhead 
> of sysadmin plus the need to find secure space for machines in a 
> crowded lab is not one to be relished. Lab space is incredibly 
> expensive too, not only to build and maintain but also because the 
> space occupied by machines is space that could otherwise house one or 
> more grant-generating postdocs :-)

Sun's price is probably only a good deal if you have a one-off task 
that needs doing.  If you need ongoing compute, it's almost certainly 
cheaper to buy your own kit (as long as you have the space).

We did some rough drunk-in-the-pub calculations based on Sun's numbers, 
and Ensembl's compute requirement.  At Sun's rates, after only 4 
months, we'd have spent as much money as our Top-500-listed cluster 
cost to buy, plus the power bill, and the sysadmins' salaries for a 
year, and probably still have change left over.

Tim

-- 
Dr Tim Cutts
Informatics Systems Group, Wellcome Trust Sanger Institute
GPG: 1024D/E3134233 FE3D 6C73 BBD6 726A A3F5  860B 3CDD 3F56 E313 4233