Author: Robert Hyatt
Date: 11:39:24 10/25/03
Go up one level in this thread
On October 25, 2003 at 14:35:37, Robert Hyatt wrote: >On October 25, 2003 at 00:15:59, Djordje Vidanovic wrote: > >>On October 24, 2003 at 20:35:24, Robert Hyatt wrote: >> >>>On October 24, 2003 at 15:45:31, Gian-Carlo Pascutto wrote: >>> >>>>On October 24, 2003 at 15:16:24, Omid David Tabibi wrote: >>>> >>>>>No threads?! How do you conduct parallel search then (if I may ask)? >>>> >>>>Multiprocessing + shared memory >>>> >>>>-- >>>>GCP >>> >>> >>>IE fork() + either mmap() or shmget()/shmat(). >>> >>>It has some drawbacks. Eugene's probe code is threaded. Not using >>>threads hurts there, as you don't get the LRU buffer management >>>efficiency in the egtb cache. It might help marginally on a NUMA box >>>to have separate copies of the code/read-only data on each processor's >>>local memory (Eugene doesn't think this is much of a factor however, >>>due to cache sizes on larger NUMA boxes). >> >> >> >>Deep Sjeng uses the ipcs command and ipcrm shm 'shmid'... for hash bigger than >>32MB. >> >>Djordje > > >You can do either one for > 32mb. mmap() or shmget. The ipcs commands are >only used to remove the things as they are persistent (shmget() shared >memory objects). Another thing... this is somewhat of a kludge. You don't need to use ipcrm command to remove a shared memory object. The program ought to do this by itself. IE set up and remove everything inside the engine...
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.