gld@cunixd.cc.columbia.edu (Gary L Dare) (11/14/88)
Re: wilson@uicbert.eecs.uic.edu (Paul R. Wilson) - In article <3508@hubcap.UUCP> you write: > >Optimistic computation is the premature computation of things that >may or may not turn out to be useful. This can increase parallelism >by allowing you to execute things without waiting to be *sure* you're >computing the right thing. > >This approach appears to work well for applications, like distributed >simulations or financial transaction mechanisms, in which most >possible interactions do not in fact happen. David Smart (now of Analog Devices) did a doctoral thesis at UIUC on parallel circuit simulation using the Gauss-Jacobi algorithm for Waveform Relaxation. I just got a copy and haven't looked at it yet, but there is a paper in the ICCD '87 proceedings summarising his work. While the superiority of this approach (which would imply optimistic computation) over Gauss-Seidel is demonstrated, I'm not sure of all the conditions imposed in the comparison or how the results would change if additional restrictions were introduced . . . gld -- ~~~~~~~~~~~~~~~~~~~~~~~~ je me souviens ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Gary L. Dare > gld@eevlsi.ee.columbia.EDU > gld@cunixd.cc.columbi.EDU "Free Canada, Trade Mulroney!" > gld@cunixc.BITNET