On Tue, 26 May 2009, Gerry Creager wrote:
How can I get sufficient cloud resources for computing... ... that will handle reasonable weather models with their small message MPI chatter, and lots of file I/O?
By asking Amazon to equip their cloud nodes with a HPC interconnect or to have fat nodes (=many cores, use only shared memory for MPI). How would you do it differently for you own cluster ? Do you really think that Amazon has the magic powder to pour over Gigabit Ethernet and turn it into 1microsecond latency interconnect ?
However, getting data into S3 for availability, when a daily multi-gigabyte dataset is used for initiation, and another is created as output, is going to be expensive, and likely slow.
I haven't tried to access S3; is the bandwidth limitation per account, per connection, per access node or some other way ? Striping comes to mind as a way to use several accounts/connections/access nodes to move data.
Another potential use: A company or research group doesn't know if they want to dip into HPC.
Argh, if they do this they might never dip into HPC ;-) -- Bogdan Costescu IWR, University of Heidelberg, INF 368, D-69120 Heidelberg, Germany Phone: +49 6221 54 8240, Fax: +49 6221 54 8850 E-mail: bogdan.coste...@iwr.uni-heidelberg.de _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
