[Gaul-devel] Storing large chromasomes with mmap
Brought to you by:
stewartadcock
From: Brian S. J. <br...@tu...> - 2002-07-24 03:13:23
|
Hi, I'm currently using GAlib for an experiment I'm running just for fun, but I might consider going with GAUL before I get too deep, since I'm definitely not a C++ afficionado to put it mildly. (Too... much... context... must... run... home... to... Mom.) This is sort of a heads-up don't-burn-the-bridge feature request. I notice you're looking to make library structures opaque, which is OK and generally a good move if sometimes done unnecessarily. However, in my case and I can imagine in many others, I'm working with large genomes which would exhaust available swap memory if I kept all the populations loaded. Thus I keep the chromasomes (the largest part) in a file on disk and mmap the file. This has a few other advantages, like saving the chromasome allele states even on ungraceful program exit. I'm still not done reading through GAUL code to see if this can be safely handled, but as development progresses I'd encourage keeping things in a state where at least the chromasome and the user's private data area are not assumed to be malloced memory, and automated memory management for those parts can be disabled. Thanks. PS. What I'm working on, just in case anyone would like to offer advice/comments is described here: http://mojo.calyx.com/~bri/writings/decay.html Though I've made more progress since writing that document and am debating whether my concept of the way the GA process should work as described there will really be a better fit for the problem than one of the stock GA's with properly chosen parameters. My main concern is that my solution space is much bigger than most people seem to be using GA's for, though not as big as it may seem since it may be relatively rich with desireable optimums. -- Brian |