Author: Vincent Diepeveen
Date: 14:18:44 05/28/02
Go up one level in this thread
On May 27, 2002 at 23:40:45, Christophe Theron wrote: >On May 27, 2002 at 18:09:25, Vincent Diepeveen wrote: > >>On May 26, 2002 at 15:06:35, Christophe Theron wrote: >> >>Yeah Christophe, we share the same viewpoint, >>but what this guy says is true. With the linux >>nerds you gotta be very careful, they are willing >>to start a holy war for it and will hate you rest >>of their life. > > >I'm well aware of this attitude and I mentionned that it is a handicap for >Linux. >That was the idea behind my "call for new blood". Another joke in the linux world is the 'progress' in linux. No unpaid dude in linux (without initially linus thorvald) allowed linux to progress. The times that linux progresses a company has paid someone money in order to work fulltime to linux. Take the improvement from gcc. Only 2 jumps in speed there were for DIEP. a big one now is from 3.0.x to 3.1 That's a COMMERCIALLY paid improvement of gcc. Some say AMD paid SUSE and/or Redhat to work on gcc. This is about the only progress that happens in linux and it is not such a nice progress. Redhat and SUSE and another few distributions who make big money on it, they work hard on linux with a few people. But it is only a few people progressing. The rest of the gcc team, about 15 guys, many whom i have chatted to either online or using email or bulletin boards, they are very capable and nice people. Yet they progress like near zero with gcc. Some idiotic professor stands up and writes down a few stupid lemma's for the kernel, which only are going to slow down the software i've written which runs under linux. Some people do not understand for example what a server is doing and what means the saying: "it gotta work bugfree and it gotta work now". Theoretic solutions regarding multithreading like: "you shouldn't poll", it's just theory. In theory in my chessprogram i should do the i/o using multithreading. However that is dead slow compared to the solution i use. I poll in windows with a function GetTickCount() in order to see whether the search times out. In windows i ask a cool guy like Scott Gasch "how many clocks is this going to cost me?" Answer is then: "it's very fast, about a 100 processor clocks or so for the NT kernel, perhaps even less". No multithreaded solution will be able to compete with that of course. already locking a variable or entity or object or even signalling an object is already taking more clocks. Yet multithreading *in theory* is much nicer to do. It is obviously far slower than polling approach. In linux i have *no clue* how many clocks i lose to a call to gettimeofday() which i use to get the wall clock time. Any 'new' blood is simply spending 99% of his time to in theory unnecessary things to find out when programming for linux. This is a basic and major problem simply. I remember my first steps in parallel programming in linux. Before that i had already run 2 processes in windows and multiprocessed it. To implement the sharing of the hashtable and some other datastructure that took me only a few hours. The total windows source code is about 20 lines of code or so. From which a few lines are windows specific code. The others checks whether i could allocate the shared memory. Everything directly worked perfect then in windows (winboard engine at the time, so ignorant from virtual adress space in these times and not needing to fix it either). To get the stuff to work in linux, just *sharing* a bit of memory > 32 MB, that took 3 months. Of course i still didn't find it out myself. Only thanks to Tim Mann and a very big hand from Bob i was able to do it. Had Bob not done so much for me, i still would be single cpu! Another thing now. I'm trying to let my interface connect to chess server. How do you disconnect in linux? Of course i know, control-c is what the USER can try himself. I don't want to do it by hand. I want the program to disconnect!! Simple questions like that. Taking days... ...in windows all you need to do is browse into msdn and DANG answer. Yet if you compare linux a few years ago, with a default redhat KDE install now, then it has gone really professional! So the progress is there. I don't wanna miss the boat, and very important, it is for free! >>I can tell you, despite using both OSes, i am >>much hated in both a windows channel saying linux >>was better in this, and i was completely banned >>out of the linux channel in IRC, when i showed a small >>thing i hated in linux. I was discussing their lack >>to fix the shared memory bugs in linux, like it >>doesn't give back the memory to the system, but >>crashes a machine instead if you control-c diep a >>few times :) >> >>Bob smartly solved this with a script that removes >>automatically the shared memory. And i still owe him >>a lot for getting out of bed in the middle of th enight >>and driving to the university, in order to reboot the >>machine, after i control-ced diep twice (2 x 400MB = 800MB >>so that crashes the kernel when the main memory is >>512MB) :) >> >>Just mentionning that they should simply implement in >>kernel what is in the 'man' pages (the _anonymous memory >>flag isn't implemented in linux, which forced me to >>use shmat/shmget functions which can crash the kernel) >>this is enough to start a holy war against a person >>if the persons in questions are linux nerds :) >> >>Bad news means signing your own death sentence. >> >>Thing is, that with all those small things which clever >>guys like Bob can fix without thinking, it is pretty >>much for the average user. >> >>But may i remind you that linux is now the fastest OS >>for diep? >> >>(of course i know i said it wrong again: GCC is fastest >>compiler for me but as it runs under linux, and it >>doesn't run for me under windows as god knows how to >>cross compile it, this means it's fastest OS for me). > > >By how much? 1, 2, 5 or 10%? 5% !!!!!! >That doesn't count much in the choice of an OS. That is *the only thing* that counts for me. >But naturally if your program can run on both OSes, then just pick the one that >gives better performance... my GUI doesn't yet. I'm busy porting it to linux now. let's see whether i can get it to run while emailing so much. > Christophe
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.