optimized applications and anonymous platform?
Message boards :
Number crunching :
optimized applications and anonymous platform?
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
Send message Joined: 18 Mar 13 Posts: 32 Credit: 2,506,320 RAC: 0 |
Last modified: 18 Aug 2013, 8:39:55 UTC Slightliy OT : I'm using an outdated core client, it doesn't report the CPU capabilities properly. So I downloaded the 32bit SSE3 version and now I'm trying with this app_info.xml : <app_info> <app> <name>period_search</name> </app> <file_info> <name>period_search_10000_windows_intelx86__sse3.exe</name> <executable/> </file_info> <app_version> <app_name>period_search</app_name> <version_num>100</version_num> <file_ref> <file_name>period_search_10000_windows_intelx86__sse3.exe</file_name> <main_program/> </file_ref> </app_version> </app_info> (Win x86 sse3 - Intel C2Q CPU) No result finished yet but I can already tell that the speed went up tremendously (estimated by factor 5). p.s.: I forgot that BOINC removes all results when you switch between BOINC downloaded application and anonymous platform (it does a complete project reset) so I lost two already started results. So before you try this yourself, crunch down your cache and report everything before switching. |
Send message Joined: 19 Jun 12 Posts: 221 Credit: 623,640 RAC: 0 |
Last modified: 18 Aug 2013, 22:58:16 UTC p.s.: I forgot that BOINC removes all results when you switch between BOINC downloaded application and anonymous platform (it does a complete project reset) so I lost two already started results. :)) No, BOINC do not do that if you do it properly (by check and copy relevant parts from client_state.xml) Your mistake is this: <app_version> <app_name>period_search</app_name> <version_num>100</version_num> (100 translates to 1.00) In my client_state.xml I find this: <app_version> <app_name>period_search</app_name> <version_num>10000</version_num> <platform>windows_intelx86</platform> <avg_ncpus>1.000000</avg_ncpus> <max_ncpus>1.000000</max_ncpus> <flops>42711869931.100609</flops> <plan_class>sse3</plan_class> <api_version>7.1.0</api_version> <file_ref> <file_name>period_search_10000_windows_intelx86__sse3.exe</file_name> <main_program/> </file_ref> </app_version> !!! If you change it now - the tasks will disappear again! But you may copy the relevant part and change only the top version (new tasks will be labeled with 100.00 , old will remain 1.00): <app_info> <app> <name>period_search</name> </app> <file_info> <name>period_search_10000_windows_intelx86__sse3.exe</name> <executable/> </file_info> <app_version> <app_name>period_search</app_name> <version_num>10000</version_num> <file_ref> <file_name>period_search_10000_windows_intelx86__sse3.exe</file_name> <main_program/> </file_ref> </app_version> <app_version> <app_name>period_search</app_name> <version_num>100</version_num> <file_ref> <file_name>period_search_10000_windows_intelx86__sse3.exe</file_name> <main_program/> </file_ref> </app_version> </app_info> - ALF - "Find out what you don't do well ..... then don't do it!" :) |
Send message Joined: 13 Mar 13 Posts: 2 Credit: 60,653,400 RAC: 0 |
Last modified: 19 Aug 2013, 2:44:12 UTC
My experience is that BOINC will delete all downloaded wu's. This is because when you first run a project normal way (without app_info.xml), BOINC will detect platform (OS + CPU architecture) and download matching application and wu's. When you use app_info.xml, platform will always be 'Anonymous', therefore BOINC will delete previous downloaded wu's (they are not supported by current platform). To my opinion this is a design flaw with 'app_info' principle; specifying platform within app_info.xml should be recognized by BOINC and work like it was specified in client_state.xml, so custom applications can process previous downloaded wu's (they are required to be compatible anyway). Also <download_url> should work, so you can distribute custom applications without the need to download them manually. |
Send message Joined: 19 Jun 12 Posts: 221 Credit: 623,640 RAC: 0 |
No. This happens only if the app_info.xml is not composed/written well. To see how app_info.xml can be written to retain all possible stock apps: Get 'Lunatics Installer' from some of the 2 links posted here: http://setiathome.berkeley.edu/forum_thread.php?id=71867&postid=1375943#1375943 Using 7-Zip unpack the file (Lunatics_Win64_v0.41_setup.exe) Look in all/any .aistub file (these are app_info.xml files with different name) Every combination of <version_num> , <platform> , <plan_class> is listed so no stock SETI task is deleted. (If you 'Add' (attach) SETI to your Windows 7 computer you can test this - after stock apps are running for a while - exit BOINC, run Lunatics_Win64_v0.41_setup.exe - choose all relevant apps (CPU, NVIDIA) in the GUI of the Installer - restart BOINC - no tasks will be deleted, started tasks will continue to run using apps from the Installer - look the auto-created app_info.xml in ...\projects\setiathome.berkeley.edu\ ('Lunatics Installer' only changes files in ...\projects\setiathome.berkeley.edu\ and nowhere else) ) - ALF - "Find out what you don't do well ..... then don't do it!" :) |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
i got a response from Thor this morning... If he does get back to you he'll only feed you more BS, eg. the guy who went on vacation went to Australia where he was eaten by a shark while surfing on the Great Barrier Reef. Frankhagen said it all. It's PNT (Polish National Team) doing what they do best. They're the best damn bunch of cheaters in BOINCdom and anybody familiar with their history knows it. I haven't figured out how they're scamming this project and I don't think I'm going to bother figuring it out but someone will, eventually. The anonymous platform thing is a red herring they use to confuse the situation by making you think they have an optimised app. They don't have such an app and the proof will be that they will never be able to produce their optimised app. It doesn't exist. Oh they'll tell you it exists but it's only for their own use so they can't share it because they want all the credits but think about it... if they've compiled an optimised app then they can surely compile a version that will cease to run after a certain date. They could share that with a few people to prove they have a legitimate optimised app without giving their opponents an appreciable advantage but I'll wager all they have is a cheat and no legitimately optimised app. BTW, I'm not calling them evil, dishonest or anything like that. I think they're very talented and I love them because the more they do what they do the sooner the rest of you fools will see this credit thing is all just a big bunch of meaningless nonsense that has never worked properly and never will. Go PNT! |
Send message Joined: 10 Jul 12 Posts: 69 Credit: 9,086,498 RAC: 0 |
|
Send message Joined: 18 Jun 12 Posts: 15 Credit: 5,027,400 RAC: 0 |
It's PNT (Polish National Team) doing what they do best. They're the best damn bunch of cheaters in BOINCdom and anybody familiar with their history knows it. I haven't figured out how they're scamming this project and I don't think I'm going to bother figuring it out but someone will, eventually. neither will i - not again. but it's most likely one of those easy things to do. they are definitely not the best but the most heinous ones. |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
Erm...sounds conclusive, but shouldn't Kyong have noticed anything? What would he have to do to notice? The answer is that he would have to do as the OP did and watch to see if any host(s) acquire a ridiculously high RAC. There might be other indicators but I don't think we can assume he is motivated to watch those indicators nor do I think we should assume he should be motivated to do so. He's busy with optimizing the app and with whatever it takes to do the research. If he's like most admins he likely doesn't give a damn about the credits and who's cheating them unless it hurts the research. IMHO, we should not assume the cheaters are returning invalid results (they do pass scrutiny of the verifier) which is to say we should not assume the cheating is hurting the research. So what should Kyong notice and how should he notice it? I have always been of the mind that an admin's most closely guarded secret should be the source code for the project's app. I can't understand why any admin/scientist would give that to anyone for it seems to inevitably lead to the same thing every time... a cadre of suspect cheaters emerges and arguments and bad feelings ensue. As for Kyong being a member of CNT, lol, I won't touch that one with a 10 foot pole except to say p is a long way from c on the keyboard. |
Send message Joined: 9 Jun 12 Posts: 584 Credit: 52,667,664 RAC: 0 |
Dagorath and frankhagen - stop offending here. For clarifiaction, I got the source codes from one member of PNT, they have them optimized for SSE2, SSE3 and AVX, I compiled them and tested. They compute fast and results are good. I was comparing results from those application to our original application and the data are ok. |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
It's PNT (Polish National Team) doing what they do best. They're the best damn bunch of cheaters in BOINCdom and anybody familiar with their history knows it. I haven't figured out how they're scamming this project and I don't think I'm going to bother figuring it out but someone will, eventually. Probably easy to do but only after you've discovered the secret. Looks to me like this is a new cheat method. Not sure but I suspect they're returning results that take less than a second to "crunch" but somehow validate. The result report spoofs the runtime (easy to do unless the server has certain patches) to make it appear as though they chewed on the task for a time short enough to make it look like they have a legitimate optimised app. Just a theory, have never done it myself but I think it could be done. How do they get results that take less than a second to "crunch"? Suppose one somehow acquires results from one's wingmen and then sends those results back to the server in a way that tricks the server into thinking the result was crunched on your own host. The 2 results would necessarily match and therefore validate. Or maybe the validator is so lax one can easily create fake results that validate? |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
Dagorath and frankhagen - stop offending here. I apologise if I've offended you or PNT, I did not intend to offend either of you. And thank you for clarifying but I still don't fully understand. You seem to be saying we are running the same optimised source as the members of PNT who frankhagen and I have accused of cheating. If that is correct then I am afraid I can't understand why their hosts seem to be crunching tasks much faster than hosts with faster CPUs. I apologise, in advance, if I've overlooked an obvious explanation but I'm heavily sedated (morphine) because I broke my shoulder a while back and there have been complications. The morphine makes me even dumber than I usually am. I could not care less who is in first place or how they acquired their credits. My only concern is the validity of the results and the progress of the investigation. |
Send message Joined: 21 Dec 12 Posts: 176 Credit: 136,462,135 RAC: 75 |
Oficial app do not use the same source as PNT have. We have decided to build our official app from scratch. I have incorporated some thoughts from PNT version especially integer comparsion which I do not believe in. Our Sse app is a little faster on same workunit and hw. We have measured both apps last two weeks. There are some compiler settings which may affect speed by +- 100s depending on cpu. We are distributing app to many different cpus so we have to choose more conservative approach. You do not see if Cpu is overclocked or not. Our megatester Nenym from CNT uses i7-3770 with times around 1100s. My i7-3770 did the same work i 1250s. Happy crunching to all guys and many thanks for contributing to our project. |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
Thanks for further clarification HA_SOFT. I understand now. When Kyong said "I got the source from one member of PNT" I assumed he meant he got the source from PNT and is using it for the app the project delivers. I apologise for the misunderstanding. To all the volunteers from PNT... my suspicions were wrong this time but there have been so many instances in the past where members of your team have cheated it's easy to suspect some of you were doing it again, not that I think it's wrong, the credits are worthless anyway so it makes no difference to me how you obtain them. |
Send message Joined: 30 Nov 12 Posts: 1 Credit: 57,168,840 RAC: 0 |
Last modified: 20 Aug 2013, 6:09:06 UTC PNT nigdy nie oszukuje. Naszym celem nie jest oszukiwanie lub destabilizacja jakiegokolwiek projektu. Naszym celem jest wsparcie projektów naukowych naszą mocą obliczeniową oraz rozległą wiedzą. PNT opracowuje własne wersje zoptymalizowanych aplikacji, kosztem znacznego poświęcenia czasu i środków finansowych, aby przeliczyć jak najwięcej zadań w jak najkrótszym czasie i zwrócić poprawne wyniki. Ma to odzwierciedlenie w rankingach i zdobytych punktach. Wysokie miejsce w rankingach jest nagrodą która nas cieszy, za trud poniesiony dla BOINC. Wszelkie pomówienia o oszustwa nie są mile widziane. Są zwykłymi oszczerstwami nie mającymi nic wspólnego z prawdą. Są drużyny które inwestują w sprzęt - są i takie, które inwestują w optymalizacje. Poznajcie prawdę a prawda Was wyzwoli. PNT - TomaszPawel P.S. Uczcie się języka polskiego - to piękny język. |
Send message Joined: 10 Jul 12 Posts: 69 Credit: 9,086,498 RAC: 0 |
|
Send message Joined: 9 Jun 12 Posts: 584 Credit: 52,667,664 RAC: 0 |
Last modified: 20 Aug 2013, 10:04:32 UTC Thanks for further clarification HA_SOFT. I understand now. When Kyong said "I got the source from one member of PNT" I assumed he meant he got the source from PNT and is using it for the app the project delivers. I apologise for the misunderstanding. I am sorry, I forgot to mention that. |
Send message Joined: 26 Jan 13 Posts: 11 Credit: 13,069,936 RAC: 0 |
Dagorath and frankhagen - stop offending here. It's good to know the results from the PNT optimized app are valid. |
Send message Joined: 16 Aug 12 Posts: 293 Credit: 1,116,280 RAC: 0 |
Thanks for further clarification HA_SOFT. I understand now. When Kyong said "I got the source from one member of PNT" I assumed he meant he got the source from PNT and is using it for the app the project delivers. I apologise for the misunderstanding. The blame is mine for reading too much into what you wrote. It worked out in the end :) |
Send message Joined: 20 Jul 13 Posts: 15 Credit: 5,985,840 RAC: 0 |
PNT nigdy nie oszukuje. translation:
|
Previous · 1 · 2
Message boards :
Number crunching :
optimized applications and anonymous platform?