4

My company is testing/comparing various grid+cloud alternatives to perform scientific calculation.
I have read the interesting white paper on the HPC/Azure topic by David Chappell, which is excellent from a "concepts" perspective, but as this is pretty new and cutting edge, I had hard times finding a complete "stupid guide" for mounting an HPC Server, putting nodes on clouds, with a Dummy "Hello HPC-Cloud World" multi-threaded .Net Program.

I've been reading/trying stuff for a couple of days and I have to admit this is a dense forest so I'm a bit lost. We can't afford allocating a resource for a whole month just for giving HPC a shot. If anybody can help on this, that will save precious time for first-time testers and adventurers like us :-)

Peace.

NB : Anticipating comments / questions asking for more precisions here : What I am ideally looking for is a how-to-guide starting from scratch, on how to mount an HPC server on-site, allocate nodes on Azure Nodes (or some other Windows Server R2 VMs ideally) (that's the cloud part), and deploy a dummy "hello-world" .Net App which have hundreds of threads that would be assigned to different nodes (that's the grid part)

Edit 1 : There is a very nice online HPC virtual lab offered by MS. Unfortunately it kept crushing on and on so we're onto finding a plain'ol "paper" guide.

Edit 2 : Listen... You here this sound ??? Listen Closer... That's the sound of the blowing wind in the desert. " Feeeeeeeeeeeeeeeeeewwwwwwwwwwwww"... See that rolling Tumbleweed there ? It's my sad and lonely question waiting for somebody to answer :-)

Mehdi LAMRANI
  • 11,289
  • 14
  • 88
  • 130
  • Well, it's been a while since you asked this question, but I would like to ask, did you ever get any real info? I am in a similar situation, trying to figure out how to get started and run the simplest of MPI/HPC jobs. The MS documentation is hopeless. – bob.sacamento May 19 '21 at 14:15

1 Answers1

1

There is a shipping version of the Microsoft HPC scheduler available @

http://www.microsoft.com/downloads/en/details.aspx?FamilyID=acde41c6-153a-4181-912e-78024fcc86da

It requires the headnode to be on-premise, but all of the workers are in the cloud. The download includes samples, etc.

There will be a new version coming (currently in beta) which will not require an on-premise Headnode (though you could still use one if you wanted to have a hybrid cluster).

Pat Filoteo
  • 1,016
  • 6
  • 4
  • Thanks. I'm ok with the on-premise HPC server as long as I get this running whithout spending months to give it a quick test. – Mehdi LAMRANI May 18 '11 at 11:53
  • This very recent document and samples are very instructive and technical (it was published just few days ago). Pretty much the kind of stuff I was looking for (hands-on code), and matches exactly our so-called "embarrassingly parallel" parameter sweeping use case. I did not stumble upon that one though I searched pretty well. Hope this post will help others. Thanx again. – Mehdi LAMRANI May 18 '11 at 12:14