cline@cheetah.ece.clarkson.edu (Marshall Cline) (05/09/90)
I'm embarassed :-) to say that I was the fool that started this lengthy `Random Points on a Sphere' discussion (sorry :-) In article <271@paralogics.UUCP> shaw@paralogics.UUCP (Guy Shaw) writes: >The thing that bothers me about using any method that involves >generating random numbers and rejecting those that fail some test >is that, even though an average rejection rate is known, there is >no guarantee that I won't generate a _long_ series of rejected numbers.... >Am I just being paranoid? ... Semi-paranoid perhaps. But for my application, Guy is right on target. My application runs on the Connection Machine, which is of course a SIMD. The problem with the `rejection' techniques is that *every* processor has to wait until *all* processors finally get an `unrejected' value. That's the price you pay to get data parallelism and a cheaper machine that doesn't have the mutual exclusion problems common to MIMD's. Marshall Cline -- =============================================================================== Marshall Cline/ECE Department/Clarkson University/Potsdam NY 13676/315-268-3868 cline@sun.soe.clarkson.edu, bitnet: BH0W@CLUTX, uunet!clutx.clarkson.edu!bh0w Career search in progress: ECE faculty. Research oriented. Will send vita. ===============================================================================