Path: utzoo!attcan!uunet!tut.cis.ohio-state.edu!cs.utexas.edu!swrinde!zaphod.mps.ohio-state.edu!usc!apple!agate!linus!linus!mwunix.mitre.org!twegner From: twegner@mwunix.mitre.org (Timothy Wegner) Newsgroups: comp.windows.x Subject: Re: MandelSpawn version 0.05 available Keywords: distributed Mandelbrot X11 Message-ID: Date: 9 Oct 90 14:07:53 GMT References: <1990Oct6.152350.26239@santra.uucp> Sender: usenet@linus.mitre.org Lines: 25 gson@joker.hut.fi (Andreas Gustafsson) writes: >The ideal environment for MandelSpawn is a workstation class with >5..50 Unix machines. A typical 1152 x 900 pixel, 250-colour image is >calculated in less than 30 seconds using 30 68030-based workstations. Sounds like a really great project! As a basis of comparison, FRACTINT on a 25mhz 80386 PC can do a 1024 x 768 Mandelbrot in 39 seconds. Not fair really, not a "typical" image - so I zoomed in about 13 times, and it took around 3 minutes. The key for you is to put in other optimizations besides parallel processing. SOunds like you may have a start since you mentioned cutting time inside the lake. The speedups we use are 1. fixed point math instead of floating point when we can get away with it 2. exploit symmetry when we can 3. detect periodicity inside the lake to bailout sooner and 4. make very educated guesses about solid areas and avoid calculations. All of these are possible for you, although #1 may not be worth it. I would say #4 is the most important because it is the most general. #2 is a non-factor for deep zooms, and #3 only helps if you are in the lake. I would be curious to know how you divide up the fractal. To use optimization #4, each CPU should do a rectangle rather than a row. You don't want the "guessing" algorithm to require network traffic. Keep it up, and good luck. Tim