Author: Scott Gasch
Date: 10:59:43 11/30/04
Go up one level in this thread
On November 30, 2004 at 02:22:49, Daniel Shawul wrote: >On November 30, 2004 at 01:31:30, Russell Reagan wrote: > >>On November 29, 2004 at 23:12:25, Daniel Shawul wrote: >> >>>I am working on a multiprocessor version of my engine >>>DanChess the previous weaks. Since i don't have a dual machine >>>to work on i need some testers. If any body is interested please >>>drop me an email. Leo Dijksmann have been testing it and found that >>>it is using 80-90% of cpu on a dual machine. So it must be really running >>>parallel. >> >> >>I can run it on my dual machine if you send me a copy. > Sent. >> >> >>>What is csrss.exe? When the multiprocessor version of my engine runs >>>this exe also runs. On the internet i see it is some file which must run >>>when a multi processor engine runs. But i am not sure if that is the case >>>for all MP engines. >> >>csrss = Client/Server Runtime SubSytem >> ^ ^ ^ ^ ^ >> >>One of its duties is creating threads. > So it doesn't matter how much cpu it uses? >Because it sometimes uses upto 20% cpu. I am not creating and destryoing lots of >search threads, 400 splits in a 10 second search maximum. > daniel How are you creating threads for your parallel search? Based on what you report with csrss.exe's overhead, I suspect you are doing something like creating / destroying threads when you split / finish a split. Everytime you create or destroy a win32 thread csrss.exe is notified via an LPC call. This is not the only overhead associated with creating / destroying a bunch of threads; think of the time it takes to allocate and free the stack commit of the new thread and the kernel data structures associated with a new thread. A better way is to create a pool of threads when the engine starts and then to use them when you need them. When you no longer need them send them back to an idle loop. Scott
This page took 0 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.