Seti@Home optimized science apps and information
Optimized Seti@Home apps => Linux => Topic started by: michael37 on 23 Apr 2009, 10:30:33 am
-
Does anyone have AP SSE2 64-bit app? Link on Crunch3r's website points nowhere, and link here doesn't work.
I have Core 2 Duo T5300 which can execute SSE3 64-bit version, but it errors out all the time. I don't know if it's a problem of AP or SSE3 app on a CPU which has limited SSE3 support. MB SSE2 64-bit works fine.
-
I've been running the sse3 64 bit app since it's release with absolutely not a single problem.
For sse2 64bit app try this:
http://calbe.dw70.de/astrop/V5/ap_503_sse2_lx64.tar.bz2
-
Thats really weird. Your cpu is most definitely a true native sse3 based cpu.
You can even run ssse3 code on it too.
Try to upgrade to latest boinc executable..
Kind regards Vyper
-
Michael, is this (http://setiathome.berkeley.edu/show_user.php?userid=11654) you? Can you give us a link to this host because they are hidden?
-
michael37, can you copy and paste one of the error messages from the failed wus on your T5300 host to this forum, please. That way we could better decide what causes the errors you see, thank you.
-
Sure, the host is here.
http://setiathome.berkeley.edu/results.php?hostid=4884128
Errors are like this:
<core_client_version>6.2.18</core_client_version>
<![CDATA[
<message>
process exited with code 193 (0xc1, -63)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
AstroPulse v. 5.03
Linux 64 bit
FFTW USE_CONVERSION_OPT USE_SSE3
SIGSEGV: segmentation violation
Stack trace (6 frames):
astropulse-5.03.x86_64-pc-linux-gnu(boinc_catch_signal+0x43)[0x4325d3]
/lib/libpthread.so.0[0x7f20d9af4080]
astropulse-5.03.x86_64-pc-linux-gnu[0x408666]
astropulse-5.03.x86_64-pc-linux-gnu[0x406536]
/lib/libc.so.6(__libc_start_main+0xe6)[0x7f20d97915a6]
astropulse-5.03.x86_64-pc-linux-gnu(__gxx_personality_v0+0x199)[0x405ca9]
Exiting...
</stderr_txt>
]]>
-
Thanks michael37. This one from your hosts resultlist is also interesting :
<core_client_version>6.2.18</core_client_version>
<![CDATA[
<message>
process exited with code 251 (0xfb, -5)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
In ap_client_main.cpp: in mainloop(): at dm_chunk_large 896
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
Error reading from statefile: wanted 13240 bytes, got 7736
</stderr_txt>
]]>
Did you change something around the 15th/16th april on your os, some updates possibly, because it looks like around that date the kind of error has changed from the above to the one you posted before. And furthermore there is a lonely completed and validated result reported today. How often does that happen ?
The kernel version 2.6.28-11-generic seems to be new and was not available when this AstroPulse app was initially tested. Maybe we have to rerun some tests with this new kernel version. Will start to look deeper into this issue to find out what is wrong, now.
-
Thanks michael37. This one from your hosts resultlist is also interesting :
<core_client_version>6.2.18</core_client_version>
<![CDATA[
<message>
process exited with code 251 (0xfb, -5)
</message>
<stderr_txt>
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
In ap_client_main.cpp: in mainloop(): at dm_chunk_large 896
In ap_gfx_main.cpp: in ap_graphics_init(): Starting client.
Error reading from statefile: wanted 13240 bytes, got 7736
</stderr_txt>
]]>
Did you change something around the 15th/16th april on your os, some updates possibly, because it looks like around that date the kind of error has changed from the above to the one you posted before. And furthermore there is a lonely completed and validated result reported today. How often does that happen ?
Ignore this error totally, that was a bug with suspend/hibernate of the laptop.
The errors that concern me are the segmentation faults.
The kernel version 2.6.28-11-generic seems to be new and was not available when this AstroPulse app was initially tested. Maybe we have to rerun some tests with this new kernel version. Will start to look deeper into this issue to find out what is wrong, now.
It is what it is, the recently released (ahem, as of today, April 23rd), version of Ubuntu 9.04 Jaunty Jackalope. Comes with new kernel, libraries, everything. Will be glad to help as always.
-
After rereading some pages of our testing forum for the current AP app i found that this same error was reported before. That person used a very old kernel version which was not fully capable/compatible with the newer features (< 2.6.22). So this could be some kind of compatibility issue. Will download the current (K)Ubuntu version over night and try out how it reacts on one of my T7200 hosts.
-
Michael as a first step, can you update your boinc to 6.6.20?
I don't think it's strictly a kernel problem, more likely a combined kernel-library (libc?) problem.
A few questions:
Were all these wus done with the same kernel? What version?
The same questions for libc6.
Were these from the prerelease ubuntu? If they were, try with the official release.
Do you have activated any power saving options (speedstep, etc.)?
EDIT: That report that Urs is referring to had also an old boinc (6.2.15, yours is 6.2.18) so again please try 6.6.20.
-
Finished testing on kubuntu 9.04 amd64 (2.6.28-11). I was able to reproduce the error michael37 did report he was seeing when using the astropulse 5.03 x64 sse3 optimized application version with Ubuntu 9.04 Jaunty Jackalope (kernel 2.6.28-11). Details of my testruns are available in the closed section.
So, if you want to use an optimized application version with kernel 2.6.28+ please use opt. astropulse 5.03 x64 sse2 application version, because astropulse x64 sse3 is not working reliable with this linux kernel.
-
Finished testing on kubuntu 9.04 amd64 (2.6.28-11). I was able to reproduce the error michael37 did report he was seeing when using the astropulse 5.03 x64 sse3 optimized application version with Ubuntu 9.04 Jaunty Jackalope (kernel 2.6.28-11). Details of my testruns are available in the closed section.
So, if you want to use an optimized application version with kernel 2.6.28+ please use opt. astropulse 5.03 x64 sse2 application version, because astropulse x64 sse3 is not working reliable with this linux kernel.
Thanks, the SSE2 app has been working for me so far.
Would be great if someone could create a Download link for it.
-
Would be great if someone could create a Download link for it.
I'll take care of this.
-
Michael do you use any boot-up options on your T5300, like acpi=off etc?
There is no general "incompatibility" of sse3 64bit and 2.6.28 kernels.
-
Michael do you use any boot-up options on your T5300, like acpi=off etc?
There is no general "incompatibility" of sse3 64bit and 2.6.28 kernels.
Nope. Acpi is working perfectly.
-
So, if you want to use an optimized application version with kernel 2.6.28+ please use opt. astropulse 5.03 x64 sse2 application version, because astropulse x64 sse3 is not working reliable with this linux kernel.
After some more testing it is time to specialize the above warning to the necessary minimum :
So, if you want to use the optimized application version astropulse 5.03 x64 sse3 with Ubuntu/Kubuntu/Xubuntu 9.04 and kernel 2.6.28-11-generic on a core 2 mobile with "Merom" core, please use astropulse 5.03 x64 sse2 optimized application version instead.
More testing has to be done to find a better solution for this problem.
-
Urs, have you tried SSE3 MB application on your computer? Looks like it might have the same type of the problem.
<core_client_version>6.4.5</core_client_version>
<![CDATA[
<message>
process exited with code 131 (0x83, -125)
</message>
<stderr_txt>
Linux optimized S@H Enhanced application by Alex Kan
Version info: SSE3 (AMD/Intel, Core 2-optimized v8-nographics) V5.13 by Alex Kan
SSE3 Linux64 Build 46 PGO, Ported by : Jason G, Raistmer, JDWhale
Processor Information:
Model: Intel Core 2 T5300, 1.73 GHz
Package: 2 Cores
Processor Caches:
L1 code cache, 32 KB
L1 data cache, 32 KB
L2 combined cache, 2 MB
Processor Features: 64bit simd [x86] cmov mmx sse sse2 sse3 ssse3 lm lahf_lm tm tm2 eist nx
Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.415339
terminate called after throwing an instance of 'std::bad_alloc'
what(): St9bad_alloc
SIGABRT: abort calledStack trace (18 frames):
AK_V8_linux64_sse3(boinc_catch_signal+0x17d)[0x4358ed]
/lib/libpthread.so.0[0x7f564ea6f080]
/lib/libc.so.6(gsignal+0x35)[0x7f564e720fb5]
/lib/libc.so.6(abort+0x183)[0x7f564e722bc3]
AK_V8_linux64_sse3[0x570c64]
AK_V8_linux64_sse3[0x565136]
AK_V8_linux64_sse3[0x565163]
AK_V8_linux64_sse3[0x56752a]
AK_V8_linux64_sse3[0x566f69]
AK_V8_linux64_sse3[0x590d47]
AK_V8_linux64_sse3[0x58e86b]
AK_V8_linux64_sse3[0x418fce]
AK_V8_linux64_sse3[0x4159cb]
AK_V8_linux64_sse3[0x414a14]
AK_V8_linux64_sse3[0x406074]
AK_V8_linux64_sse3[0x405612]
/lib/libc.so.6(__libc_start_main+0xe6)[0x7f564e70c5a6]
AK_V8_linux64_sse3(realloc+0x181)[0x405379]
Exiting...
</stderr_txt>
]]>
-
Check the new thread, its another application. First guess : Ubuntu release Jaunty Jackalope has a problem !
-
I, too, have started experiencing work unit errors with optimized astropulse when I upgraded to Jaunty. In my case almost every work unit failed (a few, say, ~10% finish successfully). Several examples:
Astropulse optimized SSE3:
http://setiathome.berkeley.edu/result.php?resultid=1212229070
http://setiathome.berkeley.edu/result.php?resultid=1213176158
http://setiathome.berkeley.edu/result.php?resultid=1213195971
http://setiathome.berkeley.edu/result.php?resultid=1213358495
Even Astropulse optimized SSE2 failed:
http://setiathome.berkeley.edu/result.php?resultid=1214211418
http://setiathome.berkeley.edu/result.php?resultid=1214221551
This is a on a Yorkfield core (Intel Core Quad Q9300). Here is a link to the machine: http://setiathome.berkeley.edu/show_host_detail.php?hostid=4600336
Note: The optimized MB SSE3 and SSE2 suffer the same fate as optimized astropulse on my system.
Other possibly relevant information:
Einstein@Home is the only other project running on this machine
VMware Server 2.0.1 is used to run a Windows XP guest
This was rock solid under Ubuntu 8.10 (i.e. never experienced a single work unit error).
I tried both BOINC 6.2.15 and 6.6.20 to see if it made any difference, but it did not.