[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Stata/MP and SSI
Henrik Stovring (firstname.lastname@example.org) asked about running Stata/MP
on a cluster computing platform using OpenSSI:
> Stata has recently announced that it ships a multi-processor version of
> its program, which is very exciting. At our system we have a so-called
> SSI cluster using OpenSSI (www.openssi.org) which is a linux
> distribution capable of turning single PC's into a single, multiple
> processor computer connected by ordinary ethernet. I would hope and
> expect that this would be an ideal environment for running Stata/MP, but
> my question is whether any have any experience or guidance on this? As I
> think this might be of general interest, I post to statalist as well in
> hope of a public answer.
Henrik also sent his question to Stata technical support, where it
arrived a few hours before it made it to Statalist. I replied privately
to him already. For others on the list who may be interested, I will
attach that reply here:
I do not believe that OpenSSI would be a good environment for Stata/MP.
Stata/MP is implemented using OpenMP and very fine-grain parallelism.
The overhead of transferring data and collecting results across a networked
grid or cluster of computers for fine-grained parallelism would far outweigh
any speed benefit from splitting the computation across multiple CPUs.
Stata/MP is only appropriate for use on a single computer with multiple
cores or processors in it.
For a cluster environment, a user could set up scripts which would launch
multiple separate instances of Stata to perform certain tasks (such as
separate parts of a simulation) and then collect the results together at
the end. We have heard of some users doing this with Stata already.
* For searches and help try: