[Bioclusters] openmosix cluster again

Joseph Landman bioclusters@bioinformatics.org
Tue, 06 Jan 2004 14:00:33 -0500


Hi Hong:

I don't think these are contradictory.  Martin has been able to use 
OpenMosix with a variety of tools.  If you note, he used the patched 
version of BLAST (which specifically disables pthreads).  The other 
tools are likely also either patched, or compiled without pthreading. 
Postgres is a database, and you really don't want it migrating from its 
data store.  Apache is a web server, and likewise, you don't want it 
migrating (extra latency etc.).

Joe

hong.zhang@research.dfci.harvard.edu wrote:

> Dear Joe and Martin,
> 
> Thanks for your instant reply. But it seems you two have contrary opinion.
> I am kinda confused. Would you please give me more explanation?
> I attached both of your answers below.
> 
> --------------------------------------------------------------------
> Martin said:
> 
> openMosix works great for our site (AR.EMBnet). We have many bio web tools
> (EMBOSS, SMS, WebPhylip) here and all of them (except SRS) work fine with
> openMosix. They migrate very well.
> On top of openMosix we use SGE for some jobs users submit by command-line.
> We use the patched version of BLAST to be able to run it (migrate) under
> openMosix.
> --------------------------------------------------------------------------
> Joe said
> 
>>   Openmosix will not migrate pthreaded jobs.  Postgres and the web
>>interfaces (apache based) are all pthreaded.
>>
>>   What open mosix is as far as bio-clusters are concerned, is a tool to
>>
>>do a "semi-automatic" load balance.  Jobs which meet specific criteria
>>can be migrated to less busy nodes.  For this to make sense, you would
>>need a) the jobs to require long running times (as compared to potential
>> migration time), b) sufficiently low IO and other resource utilization
>>such that the migration will not generate hot-spots (resource
>>contention, such as network bandwidth for accessing a file over NFS).
>>
>>   It is good for some types of jobs, though in a shared nothing
>>environment (cluster) with limited network bandwidth and relatively high
>> latencies, its isn't necessarily better than the queuing systems.
>>
>>Joe
>>
> 
> 
> 
> 
> _______________________________________________
> Bioclusters maillist  -  Bioclusters@bioinformatics.org
> https://bioinformatics.org/mailman/listinfo/bioclusters

-- 

Joseph Landman, Ph.D
Scalable Informatics LLC,
email: landman@scalableinformatics.com
web  : http://scalableinformatics.com
phone: +1 734 612 4615