Author: Vincent Diepeveen
Date: 08:49:51 10/01/00
Go up one level in this thread
On September 30, 2000 at 18:53:22, Andrew Williams wrote: >On September 30, 2000 at 17:18:59, Vincent Diepeveen wrote: > >>On September 30, 2000 at 01:28:34, Pete Galati wrote: >> >>>On September 29, 2000 at 22:42:15, Vincent Diepeveen wrote: >>> >>>>On September 29, 2000 at 18:44:13, Pete Galati wrote: >>>> >>>>>On September 29, 2000 at 17:48:15, Christophe Theron wrote: >>>>> >>>>>>On September 29, 2000 at 17:13:30, ERIQ wrote: >>>>>> >>>>>>>I for one would like to see either chess tiger II, rebel 11, or junior 6 for >>>>>>>linux or at least a good reason not to make a port like that as the unix >>>>>>>community seems to be growing and I think for the most part like chess. :) >>>>>> >>>>>> >>>>>> >>>>>>I think I could port the Chess Tiger engine to Linux in less than a week, as it >>>>>>is entirely written in C. I use the GCC compiler everyday to work under DOS >>>>>>(actually the DJGPP), so compiling to Linux would only involve system-specific >>>>>>issues, probably no compiler specific issues. >>>>>> >>>>>>The Tiger engine has been recently ported to another system running a different >>>>>>processor family than the x86, so I guess I have solved most of the portability >>>>>>issues by now. >>>>>> >>>>>>The only problems I see are commercial ones. Are there enough people ready to >>>>>>pay for Linux software? Is the system going to be improved and more user >>>>>>friendly? Are there good copy-protection schemes for Linux? >>>>>> >>>>>>When we get a positive answer to all these questions, then we will certainly go >>>>>>for Linux. >>>>>> >>>>>>I know the system is evolving quickly, and a lot of energies are now focused on >>>>>>it. I hope Linux will offer a serious alternative to Windows in the near future. >>>>>>Wait and see... >>>>>> >>>>>> >>>>>> >>>>>> Christophe >>>>> >>>>>If I were you (I'm not) I would consider approaching it like what was done with >>>>>Gandalf for Winboard, only do a Chess Tiger for Xboard. You'd be the first kid >>>>>on the block to offer a commercial Xboard engine (for whatever that's worth) >>>>> >>>>>Pete >>>> >>>>wrong, you can buy diep from me for xboard. it works perfect, >>>>see how it runs like the sun under fics. >>> >>>As usual, I didn't check my facts. Are you using a Linux version of Diep in >>>tournaments? >> >>Bob regurarly offered DIEP to run at a quad xeon for several tournaments, >>then i go for linux obviously. >> >>Now i have myself a cool system: dual PIII800, diep is 10.8% faster >>with the msvc compiler which only works of course under windows, as >>with the gcc compiler at a PIII. At PII the difference is less, more >>like 8.5%, not to mention also that a PIII itself is over 17% faster >>as a PII, Xeon is 7% faster as a PII. >> >>So a quad xeon would be speedup * 0.9 * 0.9 * 550 = speedup * 445.5 >> a dual 2.0 * 1.0 * 1.0 * 800 = 1600 >> >>So to get a quad xeon 550 as fast as a dual PIII800 for diep then i >>would need to get a speedup of 3.6 >> >>Nevertheless the advantage of a local machine without internet >>problems is always a big one, and a quad is always outperforming >>a dual somehow. It *does* get that nodes a second. Even if it >>searches complete bullshit with 1.5 processor, then still it >>is filling hashtable which can be used later. >> >>So despite the theoretical equalness of a quad to a dual for DIEP, >>the quad always outperforms the dual practical, basically >>because of the better working hashtables, not to mention the fact >>that my harddisk is dead slow! >> >>The huge disadvantage is that it runs remote, in paderborn, both >>at ipcc and especially at wccc i wasn't able to connect to the usa >>at all (wccc99 2 out of the 7 games!), >>as in germany crucial nodes that are between netherlands and germany >>seem to turn off their computer in weekends. >> >>This weak internet in europe at least and especially the huge lag >>to the USA is very frustrating. >> >>Yet it's obvious that running under linux has more advantages as just >>making use of the generous offer from Bob! >> >>One of the big advantages is that you have more test possibilities, >>like there are many commercial boundscheckers nowadays for windows, >>but there is a great free working boundchecker for gcc 2.95.2, which >>i use nowadays regurarly to check DIEP everywhere out. >> >>Also the fact that it's the only serious competitor for the PC with >>windows is an important issue. If there would be other OSes as windows >>i doubt i would have done that much effort for linux! >> >>Yet one thing that sucks bigtime is graphical development under X windows. >> >>There is nothing to nil for X windows to use in an easy way. >> >>I am NOT gonna make a diep graphical interface under linux at all actually >>without real big commercial interests, the possibility to sell a version or 2 >>is not what is tempting to do a big job like that. >> >>What you need is a good cross development platform. There is nothing working >>well in this area. >> >>I tried Qt, WxWindows i tried really hard, and some other stuff. tcl/tk i >>already tried years ago. >> >>It all sucks bigtime. I develop in first place for windows, linux is second. >> >>So basically only my engine works under linux. The whole interface is >>all win32 API-C SDK code. >> >>>>but sure christophe won't get rich. i doubt he'll sell more as >>>>a single copy. for sure he won't sell it to the 'pete' here. >>> >>>Oh, you mean me, or the several people here that stole my first name? Yeah, >>>it's hard to say, Ed's company does a good job of marketing Chess programs, but >>>you're not really set up for doing that yourself. Does anyone other than >>>yourself sell Diep, such as ICD, or Gambitsoft? >> >>Not at the moment, but what do i have to sell except a tough to install >>winboard version? >> >>Not to mention the xboard version. No one ever gets that to work >>without learning first a bit about shared memory from linux. >> > >Are you talking about your program or programs in general? Mine >works fine and I know nothing about shared memory in Linux. > >>If you by accident kill diep somehow you need to use >>(or use shell script performing) commands like >> ipcs >> ipcrm shm <segment number> >> >>If you don't do that, and diep is control-c-ed twice, then linux >>is hung. Only getting out power cable works then. >> > >I don't understand this at all. Is it a consequence of your program >being SMP? i allocate shared memory. shared memory in linux isn't returned to the system if the process is no longer running that allocated it. instead it might get 'root' rights, and if the total amount of shared memory is something like a bit bigger as your RAM+swap then your system is hung. Can be that it just needs to be a bit bigger as 127mb+RAM too didn't test the exact amount needed to crash, but something bigger as what you have RAM. So if i allocate at my dual 800 about 200mb ram, i control-c then i again start diep and control-c it, then machine is hung. of course if i set hashtable to a small value i need to control-c it more. basically it's even so that i do not allocate shared memory even. i allocate a few IPC_PRIVATE segments. Linux doesn't have anonymous shared memory anyway. Not implemented in kernel (!!!!), but there is a known bug in linux that allows more as 1 process to attach to segments that are allocated as private memory. Don't post this to the SMP kernel though. People there are not responsible seemingly for this part of the kernel. Cox isn't at least. the lord knows who is... ...this issue gets each year posted in the mailing lists each year since 1997 already, but it seems that they don't need or don't want to implement shared memory better in linux. Note that when talking about parallellism, that windows isn't holy either, but that's a different 1000 page story. Basically both OS-es are completely sequential in their design, so that's the real problem when talking about shared memory and linux. >Andrew
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.