21)
Message boards :
Sixtrack Application :
More Work
(Message 32479)
Posted 20 Sep 2017 by Eric Mcintosh Post: New work is being submitted and appears to be taken! Eric 12:28 CST |
22)
Message boards :
Sixtrack Application :
More Work
(Message 32478)
Posted 20 Sep 2017 by Eric Mcintosh Post: I am trying to submit new work, but there seems to be a network/AFS/CERN service problem. More work as soon as possible. Work has been pulled because of some of it was incorrect due to changes required by the new SixTrack executable in the BEAM block of fort.3 input. The tasks I know about all had names beginning Jone.....but that was only about 3,500 and I deleted the WUs. More news soonest. Eric. |
23)
Message boards :
Sixtrack Application :
Sixtrack tasks with name starting with Jone0_sep_jta_bbo are crashing on my system
(Message 32461)
Posted 18 Sep 2017 by Eric Mcintosh Post: Dear All, apologies for the "bad" WUs. I have tried to delete them myself. Eric. Well I have tried to delete all the "bad" Jone WUs. I might even have succeeded. I can do no more for now. Kyrre should please check the BEAM block and let Dobrin know. It looks wrong to me with 7 items when only 4 or 5 are expected........ MYSQL does not like an "_" in the names! It has a special significance. (I'll now have to print another giant manual for MYSQL.) The boinc ops facility for deleting WUs then fails because of the "_"s in the WU name. Luckily it turns out that just "Jone" and not "Jone_" was sufficient to select the WUs. Afrer double_checking I see all such Jone WUs have server_state 5 which means "OVER" although some have (of course) outcome 4 "IN PROGRESS". I think that means they will be deleted in the normal way. |
24)
Message boards :
Sixtrack Application :
Sixtrack tasks with name starting with Jone0_sep_jta_bbo are crashing on my system
(Message 32460)
Posted 18 Sep 2017 by Eric Mcintosh Post: Thanks a lot; chasing it with colleagues. Either a bug in new SixTrack or a user error. Eric. +1 |
25)
Message boards :
Sixtrack Application :
260.000 WUs to send, but no handed out
(Message 32451)
Posted 17 Sep 2017 by Eric Mcintosh Post: There is a major problem with the distribution of tasks (dating from the integration of the other projects with SixTrack). I myself am suffering like you. My 4 processor Windows 10 couldn't get work either. I tried reset, tried a re-install of the BOINC manager, etc and finally I got 100 Tasks! I was now using the client defaults, although I normally don't cache tasks. So it seems you either get no tasks or too many. The problem is being looked at but no solution seems to be available. I agree the VirtualBox message is annoying but the IT emphasis is on trying to get more clients running for the other sub-projects. In addition the BOINC project management pages are so slow to access, several minutes for each click, as to be unusable. This makes it very difficult for me to follow up on other problems. I think we still have some HyperThreading/ BIOS problem, process creation failing on Windows, tasks not being distributed. The slow start up on Windows 10 seems to be solved with the new executables, which seem to be running well. The overall error rate is down to about 2.5%, but we can do better. Apologies and I await action from support. Eric. |
26)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31900)
Posted 9 Aug 2017 by Eric Mcintosh Post: Sorry I can'h help with this. I'll pass it on. (We have been having upload, even download, problems for quite some time now, like years! I have made many suggestions but await a fix.) Maybe you could turn on some file transfer debug options for the event log and send it to eric.mcintosh@cern.ch. Eric. This WU say uploading, but don't do this: |
27)
Message boards :
Sixtrack Application :
SIXTRACKTEST
(Message 31884)
Posted 7 Aug 2017 by Eric Mcintosh Post: Thanks a lot; I was about to ask if the "real time" problem had gone away on Windows 10! Great news. Eric. (More news tomorrow.) I got several sixtracktest tasks on different systems: Linux64, Win7 64bit, Win10 32bit. |
28)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31863)
Posted 7 Aug 2017 by Eric Mcintosh Post: Well, you don't have an SSE2 version! I KNOW you should have.... I just wonder if this isn't some manifestation of a very old ifort issue where Intel wouldn't use SSE2 on AMD hardware....but since this is non-SSE version........there must be some kind of emulation... memory required is about 64MB. Eric. sixtrack-production task with 243k seconds duration with Linux (451.07)! |
29)
Message boards :
Sixtrack Application :
SIXTRACKTEST
(Message 31862)
Posted 7 Aug 2017 by Eric Mcintosh Post: Very good; your comment on memory is correct. My colleagues are trying to use a "big" version to limit the total number of versions. I am hoping that the working set size will be as before and that there will not be excessive paging. Eric. had about 60 sixtracktest, all still waiting for validation (wingman), but no errors were reported (on windows and linux). |
30)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31858)
Posted 7 Aug 2017 by Eric Mcintosh Post: Testing so far; actually have 2 validated results, but 17 errors. The errors look like infrastructure, many can't download the input files or "couldn't start app: CreateProcess() failed - " on Windows. Eric. |
31)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31856)
Posted 7 Aug 2017 by Eric Mcintosh Post: The 32-bit versions you quote are for production not test??????? Eric. tullio wrote:I am running the 64-bit app sse2 on my Linux laptop. 4 are waiting validation, 1 64-bit and 3 32-bit. |
32)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31855)
Posted 7 Aug 2017 by Eric Mcintosh Post: Well may you ask! I hope we shall release in a few weeks (best guess, and I am in principle an optimist). We are currently testing extensively on "sixtracktest". I myself submitted a few thousand cases yesterday. The current (not yet final) version under test is v46.15 as against v451.07 from 2014! The first problem was the end of Windows XP at CERN. I could never build on Windows 7, never mind Windows 10, using cygwin. My young colleagues now have a superb build system which includes Macs, ARM, and IBM P8/9. Subsequent problems with gcc and ifort have been solved, and I myself am now optimising to get the last few percent of performance. We are very eager to introduce as there is a lot of new "physics" and functionality. However we must test very thoroughly as there are millions of hours of your CPU time at stake. There are also far too many infrastructure problems, not forgetting the BIOS/Hyper Threading problem, which might go away :-) as we are now using gfortran rather than Intel ifort. More news when I can, along with some report on testing progress. Eric. The current list of available executables is (but I see my old friend 32-bit Pentium 3/no SSE is missing): sixtracktest_aarch_lin64_4615_avx.linux: ELF 64-bit LSB executable, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, not stripped sixtracktest_aarch_lin64_4615_sse2.linux: ELF 64-bit LSB executable, version 1 (GNU/Linux), statically linked, for GNU/Linux 3.7.0, not stripped sixtracktest_darwin_4615_avx.exe: Mach-O 64-bit executable sixtracktest_darwin_4615_gen.exe: Mach-O 64-bit executable sixtracktest_freeBSD64_4615_avx.exe: ELF 64-bit LSB executable, x86-64, version 1 (FreeBSD), statically linked, for FreeBSD 11.0 (1100122), FreeBSD-style, not stripped sixtracktest_freeBSD64_4615_sse2.exe: ELF 64-bit LSB executable, x86-64, version 1 (FreeBSD), statically linked, for FreeBSD 11.0 (1100122), FreeBSD-style, not stripped sixtracktest_lin32_4615_sse2.linux: ELF 32-bit LSB executable, Intel 80386, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.32, not stripped sixtracktest_lin64_4615_avx.linux: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.32, not stripped sixtracktest_lin64_4615_sse2.linux: ELF 64-bit LSB executable, x86-64, version 1 (GNU/Linux), statically linked, for GNU/Linux 2.6.32, not stripped sixtracktest_netBSD64_4615_avx.exe: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, for NetBSD 7.0.2, not stripped sixtracktest_netBSD64_4615_sse2.exe: ELF 64-bit LSB executable, x86-64, version 1 (SYSV), statically linked, for NetBSD 7.0.2, not stripped sixtracktest_win32_4615_sse2.exe: PE32 executable for MS Windows (console) Intel 80386 32-bit sixtracktest_win64_4615_avx.exe: PE32+ executable for MS Windows (console) Mono/.Net assembly sixtracktest_win64_4615_sse2.exe: PE32+ executable for MS Windows (console) Mono/.Net assembly Ah... the mention of new 64 bit apps ? What we are running now for 6 Track is all from 2014. |
33)
Message boards :
Number crunching :
Getting mostly Six Track tasks
(Message 31843)
Posted 6 Aug 2017 by Eric Mcintosh Post: Good, excuse my ignorance but what did you reset to 0? Thanks. Eric. Checklist 4 is double green. |
34)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31840)
Posted 6 Aug 2017 by Eric Mcintosh Post: Well it should NOT be necessary to eliminate the non-SSE. I pride myself on being all inclusive. It should be better very soon as the 64-bit versions will be SSE or AVX only. As of today we just run 32-bit versions on all platforms. I'll try and have a look myself. There is NO logical reason to run the "old" non-SSE versions except on ancient P3 etc! Eric. That's terrible! This was supposed to be solved. |
35)
Message boards :
Number crunching :
Getting mostly Six Track tasks
(Message 31839)
Posted 6 Aug 2017 by Eric Mcintosh Post: Well there is of course work available for both SixTrack and the experiments! Just another manifestation of the scheduler problems with KHC@home. Apologies but I can't do anything. Eric. Vbox was out of date, installed new version. |
36)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31806)
Posted 4 Aug 2017 by Eric Mcintosh Post: That's terrible! This was supposed to be solved. I'll report it AGAIN! Eric. Duration-time 13k seconds (Windows) against 160k (Linux): |
37)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31799)
Posted 4 Aug 2017 by Eric Mcintosh Post: The 5 wingman tasks exist but are unsent. Eric. My list of sixtracks with status "validation pending" show a couple of older WUs (since 2017-07-04) that have not yet been sent to any wingman. |
38)
Message boards :
Sixtrack Application :
"SixTrack Tasks NOT being distributed
(Message 31791)
Posted 2 Aug 2017 by Eric Mcintosh Post: There are now at least 2 of us who see this as a REAL problem, at least for SixTrack. Hopefully we can pin it down and fix. Eric. |
39)
Message boards :
Sixtrack Application :
SIXTRACKTEST
(Message 31790)
Posted 2 Aug 2017 by Eric Mcintosh Post: Great; I can now run on sixtracktest thanks to my colleague Alessio. There will be many more to come as I need statistics and error analysis for my upcoming paper. Just meed to check results first. Eric. I just received 15 new sixtracktest tasks, they went back in the queue behind 17 sixtrack tasks. Let's see how they perform. |
40)
Message boards :
Sixtrack Application :
exceeded elapsed time limit 30940.80 (180000000.00G/5817.56G)
(Message 31785)
Posted 2 Aug 2017 by Eric Mcintosh Post: A fix has been applied to the Validator to redefine outliers. Seems to be working as I see the overall error rate has dropped significantly. If you still the REAL_TIME_EXCEEDED please let me know. If you have a hyper fast machine maybe it is enough to rerun the benchmark on the BOINC client i.e. your side. Eric. |
©2024 CERN