Parallel computing

Message boards : Number crunching : Parallel computing

To post messages, you must log in.

AuthorMessage
Profile Emigdio Lopez Laburu

Send message
Joined: 25 Feb 06
Posts: 61
Credit: 40,240,061
RAC: 0
Message 76003 - Posted: 31 Aug 2013, 10:46:35 UTC
Last modified: 31 Aug 2013, 11:26:23 UTC

Hi, all.

I,m thinking to "build" a parallel supercomputer to run Rosetta.

I,ve discovered "http://www.parallella.org/" which is an small supercomputer BUT with ARM architecture. It runs Ubuntu BUT is ARM. Do you think it will be able to run Rosetta??. The answer, perhaps, is "no"... :( because we need an x86, correct?.

Do you know any other alternative?.

Thanks in advance.
ID: 76003 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile dcdc

Send message
Joined: 3 Nov 05
Posts: 1831
Credit: 119,627,225
RAC: 9,274
Message 76005 - Posted: 31 Aug 2013, 11:13:13 UTC
Last modified: 31 Aug 2013, 11:22:02 UTC

Hi

You are correct - the answer is no because there is no ARM port for Rosetta.

Rosetta is (and all distributed computing projects are) already massively parallel projects so having fast interconnects between the nodes doesn't help (although if it were common then it might be useful, but would need programming for).

Also, calling it a "supercomputer" is a bit of a stretch - that setup is much slower than a typical PC by almost any metric from what I've read so far. Also, Rosetta would be hit more than most benchmarks becuase the CPUs don't appear to have any cache.

Basically, it might be great for playing around with, and some parallel applications, possibly in a very low power budget, but I expect a £50 Intel/AMD CPU would be much faster for Rosetta even if there were an ARM port.

The only way you can practically do something similar for Rosetta is to run more PCs, and the cheapest/most efficient way to do that would probbaly be a stack of caseless low end motherboards/CPUs, but that's not cheap!

Additionally, you might find a boost from increasing the run-time of your tasks - there is some overhead at the start of each task which is reduced by running them for longer, so increasing your run-time preference from 6-hours to 24 might give you a boost in througput without any cost.

HTH
Danny
ID: 76005 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : Parallel computing



©2024 University of Washington
https://www.bakerlab.org