[Bioperl-l] Re: [PBS-USERS] openPBS configuration (fwd)

[Chris Dagdigian]

Hi folks-

I've worked quite a bit with LSF, PBS, PBSPro and Sun GridEngine over 
the last 2 years or so. Actually most of the paid work (ie non-OpenBio) 
work that I've done for the last 12 months has been building large linux 
clusters for bioinformatics.

IMHO Platform LSF is by far the best distributed resource management 
suite out there. The reason I say this is because it has a far lower day 
to day administrative burden and much better resiliancy and fault 
tolerance features than any other system out there. In particular the 
fault tolerance features of LSF blow all the others away.

It is however, very expensive :) Last time I checked they were still 
selling academic licenses for roughly $400 USD per Linux CPU.

At a basic level any of the DRM suites will get the job done (ie remote 
execution and batch scheduling facilities). The differences between the 
products are mostly due to scaling ability and ease of management.

The cost is still worth it for large production systems where you don't 
want to have to hire an additional expensive IT staffer to keep your 
batch scheduler from crashing and losing jobs and data every day.

Regarding PBS. I agree with your sentiments regarding the open source 
free version of OpenPBS. Unless you are willing to dive into the source 
code, apply patches and generally screw around deep in the internals of 
the code I don't think that OpenPBS is all that suitable for people 
running out of the box. It requires time and expertise that you really 
should not have to apply to a basic software layer.

On the plus side I really like and respect the people at Veridian 
systems who manage both OpenPBS and PBSPro. The commercial version of 
PBS ('PBSPro') seems to be far better than OpenPBS. The pricing is 
reasonable, you can get expert consultants and professional services to 
help with configuration and tuning issues _and_ they are willing to give 
you the source to the Pro product if you needed it.  PBSPro is also free 
to academics.

There is an alternative you may want to look at:

Sun GridEngine

Sun GridEngine was formally known as a product called 'codeine'. Sun 
bought the product and has committed to making it a key part of their 
grid computing efforts. The product is open sourced and available for 
free at http://gridengine.sunsource.net

I place GridEngine in the middle of the spectrum between LSF and 
OpenPBS. The linux version seems to work quite well for what I've needed 
to use it for and it is _rapidly_ improving. I think they will be in a 
position to eventually challenge the LSF Base functionality although 
Platform is working on lots of advanced stuff that will probably keep 
them succesfull.

I wrote up my experiences on installing GridEngine on a 64-CPU cluster 
at Steven Brenner's lab. It can be found online at
http://bioteam.net/dag/biocluster-diary-1/


That page is basically the first of what I hope to be many 
pseudo-whitepapers that are publicized on the 
bioclusters@bioinformatics.org mailing list.

Speaking of that list. This is a good time to advertise it here. If 
there is anyone reading this message who is at all interested in 
distributed computing for the life sciences then you should really join 
in on the "bioclusters" mailing list. You can subscribe online at 
http://www.bioinformatics.org/mailman/listinfo/bioclusters

If cluster hardware is a fetish you can see pictures from the recent 
comput farm projects I've worked on at: 
http://gw.sonsorol.net:8080/gallery/bioclusters


On a side note I've been reading the pipeline discussion threads with 
interest and would like to become involved if I can. I was already 
thinking about trying to make sure it plays nicely with gridengine :)

Just my $.02

Regards,
Chris

http://bioteam.net



Elia Stupka wrote:

> Hello folks,
> 
> I am pretty disappointed. Thought that PBS would be a nice free
> alternative to LSF for those who don't want to spend on LSF, but from the
> e-mail below on the PBS-users list it is absolutely obvious that their
> strategy is that PBSpro (read paid software) will be reliable, robust and
> scalable, while the free counterpart will be left unscalable and buggy...
> 
> Questions:
> 1)Do people have better info then my mail-sniffing to judge?
> 2)Is there yet another open source alternative to PBS and LSF?
> 
> Elia
> 
>