Use distributed memory depending on your plaworm you

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: nection network not specified o༆  λ is not specified beyond λ>>1 -- cannot be because every machine is different o༆  Controller, combining network “optional” Source: Larry Snyder, CSEP524, UW Discuss logP paper here Closing note on logP •  Refinements have been proposed over Ime –  “Oh, if we also measured xyz, the model would be bemer!” –  e.g., logGP: takes “long messages” into account –  Challenge: when to stop? CSEP 524: Parallel ComputaIon Winter 2013: Chamberlain 63 Abstract Machine Model Summary: My Take PRAM: Too abstract/unrealisIc! logP: Too parameterized (?) –  or, perhaps most appropriate for low- level library writer CTA: The goldilocks soluIon? By analogy: Consider sequenIal programming in a cache- aware manner without worrying about an architecture’s precise… …latencies to access different levels of cache/memory …cache lines sizes …etc. It can yield a significant fracIon of peak performance, while remaining quite portable CSEP 524: Parallel ComputaIon...
View Full Document

This document was uploaded on 04/04/2014.

Ask a homework question - tutors are online