Thanks Nicolas and Anand. My first idea was just to run the starcluster AMI, but maybe more could be done with Eucalyptus.
This is more a medium-term idea, so when the priority stuff get settled a little bit or this got more priority, I will definitely test this. Fred On Wed, Dec 15, 2010 at 5:28 AM, Anand Patil <anand.prabhakar.pa...@gmail.com> wrote: > It does look an awful lot like these iso > images: http://web.mit.edu/stardev/cluster/download_amis.html are meant to > be used on any old Ubuntu box. Is that correct? If so, I'd be interested in > learning how they did that. > Anand > > 2010/12/14 Nicolas Pinto <pi...@mit.edu> >> >> Fred, >> Not entirely sure it will work "as-is", but you can look at Eucalyptus: >> http://www.eucalyptus.com/ >> starcluster has been shown to work with it, eg: >> http://mailman.mit.edu/pipermail/starcluster/2010-March/000095.html >> >> HTH >> N >> 2010/12/14 Frédéric Bastien <no...@nouiz.org> >>> >>> Just a quick question that google did not give me an answer. >>> >>> Is it possible to use amazon AMI on our own computer? >>> >>> We have access to many cluster with different configuration, we could >>> use the same configuration in that case(module some work around >>> optimized blas). In addition, if we want to use amazon before a >>> deadline, this would still use the same computing environment. >>> >>> Thanks >>> >>> Frédéric Bastien >>> >>> On Tue, Dec 14, 2010 at 1:13 PM, Josh Bleecher Snyder >>> <joshar...@gmail.com> wrote: >>> > On Tue, Dec 14, 2010 at 1:01 PM, Anand Patil >>> > <anand.prabhakar.pa...@gmail.com> wrote: >>> >> On Tue, Dec 14, 2010 at 5:31 PM, Nicolas Pinto <pi...@mit.edu> wrote: >>> >>> >>> >>> Dear Anand, >>> >>> FYI, Justin Riley (cc'ed) already created an AMI with many of the >>> >>> libraries you listed (including pycuda, pyopencl, atlas-enabled >>> >>> numpy/scipy >>> >>> and many more) for his *wonderful* starcluster project and we'll also >>> >>> use >>> >>> them for the upcoming "Massively Parallel Computing" at Harvard >>> >>> (cs264.org). >>> >>> HTH >>> >>> N >>> >> >>> >> Thanks for the tip! StarCluster does indeed look terrific but I >>> >> couldn't >>> >> find the AMI for cg1.4xlarge in either the distribution or Google. >>> >> Justin, I >>> >> would be very keen to learn more about the AMI either on this list or >>> >> offline. >>> > >>> > Just so you know, the missing AMI is on their radar -- see >>> > https://github.com/jtriley/StarCluster/issues#issue/9. FWIW, this was >>> > also discussed on the theano mailing list >>> > >>> > (http://groups.google.com/group/theano-users/browse_thread/thread/db0f1bdc6076014e). >>> > >>> > -josh >>> > >>> > _______________________________________________ >>> > PyCUDA mailing list >>> > PyCUDA@tiker.net >>> > http://lists.tiker.net/listinfo/pycuda >>> > >>> >>> _______________________________________________ >>> PyCUDA mailing list >>> PyCUDA@tiker.net >>> http://lists.tiker.net/listinfo/pycuda >> >> >> >> -- >> Nicolas Pinto, PhD >> Research Scientist in Brain & Computer Sciences >> The Rowland Institute at Harvard, USA >> Massachusetts Institute of Technology, USA >> http://web.mit.edu/pinto >> >> >> _______________________________________________ >> PyCUDA mailing list >> PyCUDA@tiker.net >> http://lists.tiker.net/listinfo/pycuda >> > > _______________________________________________ PyCUDA mailing list PyCUDA@tiker.net http://lists.tiker.net/listinfo/pycuda