Comments on: Alternative delivery models for HPC? http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/ HPC News Without the Noise for Supercomputing Professionals | insideHPC Sun, 09 Jun 2013 01:54:13 +0000 hourly 1 http://wordpress.org/?v=3.1.1 By: John West http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24962 John West Thu, 03 Jan 2008 22:03:14 +0000 http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24962 Erik - all bets are off once you get to classified and compartmentalized computing. I'm really here talking about routine requirements and non-classified R&D. The NNSA has a clear argument against outsourcing HPC, but its not clear that other federal agencies currently building out huge infrastructures (NASA, NOAA, and others) do. Certainly universities and enterprises doing research with HPC, rather than research on HPC, would qualify as well. Erik – all bets are off once you get to classified and compartmentalized computing. I’m really here talking about routine requirements and non-classified R&D. The NNSA has a clear argument against outsourcing HPC, but its not clear that other federal agencies currently building out huge infrastructures (NASA, NOAA, and others) do. Certainly universities and enterprises doing research with HPC, rather than research on HPC, would qualify as well.

]]> By: Erik Paulson http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24957 Erik Paulson Thu, 03 Jan 2008 21:32:47 +0000 http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24957 John - Where do you work? Some of the places that need an Act of Congress to get a new building to house their new machine have very convincing arguments about why they can't ever let their code and data onto even the unclassified machine room in the same building. John -

Where do you work? Some of the places that need an Act of Congress to get a new building to house their new machine have very convincing arguments about why they can’t ever let their code and data onto even the unclassified machine room in the same building.

]]> By: John West http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24030 John West Thu, 27 Dec 2007 19:38:22 +0000 http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-24030 Deepak - I think that's a good point. If we're outsourcing all of this to "professionals" then one of the benefits we should get is better uptime. Although all of us that run centers would like to think of ourselves as professionals, the fact is that compared to the very large internet content providers we are just hobbyists when it comes to provisioning always-on resources. A company that specializes in HPC hosting should be able to have everything available all the time. Deepak – I think that’s a good point. If we’re outsourcing all of this to “professionals” then one of the benefits we should get is better uptime. Although all of us that run centers would like to think of ourselves as professionals, the fact is that compared to the very large internet content providers we are just hobbyists when it comes to provisioning always-on resources. A company that specializes in HPC hosting should be able to have everything available all the time.

]]> By: Deepak http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-23921 Deepak Wed, 26 Dec 2007 18:24:04 +0000 http://insidehpc.com/2007/12/26/alternative-delivery-models-for-hpc/#comment-23921 Actually I believe that except in certain cases it is inevitable. I've heard people from Amazon use the phrase "we take care of the muck" to describe their web services infrastructure. This is not too different from that. As networking bandwidths improve, why not. However, unlike traditional supercomputing centers, resources should always be available and only the largest jobs should be queued up. Actually I believe that except in certain cases it is inevitable. I’ve heard people from Amazon use the phrase “we take care of the muck” to describe their web services infrastructure. This is not too different from that. As networking bandwidths improve, why not. However, unlike traditional supercomputing centers, resources should always be available and only the largest jobs should be queued up.

]]>