[Bioclusters] cluster or SMP

[Robert Olson]

Speaking of diskless clusters, we actually netboot our Mac cluster  
(we run Debian), but have a 500G drive installed on each node that we  
can wipe at will  without affecting the OS. This has worked out well  
for us. The netboot makes administration of the cluster nearly  
completely painless, but we get the good performance of local I/O.

--bob

On Sep 4, 2007, at 11:46 AM, Chris Dagdigian wrote:

>
> Diskless clusters are generally a bad idea for any system that will  
> be running life science informatics workflows. In general terms  
> (without knowing your specific applications) it is generally a rule  
> that most life science codes are performance bound by memory and  
> disk access I/O limitations. If you build a diskless cluster and do  
> lots of assemblies on genomes that are larger than what can fit in  
> memory your cluster is going to be bogged down while nodes wait on  
> network I/O requests.
>
> You can always tell a HPC vendor who knows nothing about life  
> science because they try to sell you (a) MPI-optimized "beowulf"  
> systems or (b) diskless systems, both of which are totally (in most  
> cases) inappropriate technology choices for our styles of work.
>
> I think you are on to something with the SMP idea. You should  
> research how many CPUs and how much memory you can pack into a  
> single chassis along with a good performing storage subsystem.
>
> My $.02 of course !
>
> Regards,
> Chris
>
>
> On Sep 1, 2007, at 3:56 PM, Marcos de Carvalho wrote:
>
>>
>>
>> Hi list,
>>
>> I am in charge of the setup of a new bioinformatics lab in my  
>> university. However, I am wondering what would be the best for my  
>> current budget for the high-performance machine, that will do  
>> basically genome assemblies, gene finding and homology searches  
>> for at least 3 metagenomics projects (but it is probably that it  
>> will be used for other applications, like molecular dynamics). For  
>> this machine specifically I have about 23 000 US$ and my first  
>> though was to build a cluster (which at current local prices could  
>> give me a 32 node diskless beowulf with E4400 chips or 16 nodes  
>> with Q6600 chips). However, I saw some pretty good SMP machines  
>> (with a max of 8 dual core opterons), for about the same price.
>>
>>  Taking off the fun of building the cluster, the relative easier  
>> administration of a SMP machine and their more general purpose  
>> application could justify their choose over a cluster? Even with  
>> 48 more cores, could the network be a serious bottleneck in  
>> comparison with the SMP machine? Does gigabit port trunking could  
>> be a solution for the network bottleneck?
>>
>>  I am not in a hurry for this machine, so the time of building the  
>> cluster can be discarded.
>>
>> Thanks in advance.
>>
>> Regards,
>>
>>
>> Marcos
>> _______________________________________________
>> Bioclusters maillist  -  Bioclusters at bioinformatics.org
>> https://bioinformatics.org/mailman/listinfo/bioclusters
>
> _______________________________________________
> Bioclusters maillist  -  Bioclusters at bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters
>