48 cores linux64

Author	Message
Mugurel Send message Joined: Jan 20 06 Posts: 3 Credit: 17,799,195 RAC: 10,407	Message 111036 - Posted 10 Mar 2011 13:28:24 UTC
Hi, Einstein@home binaries would fail to run on my new 48cores nodes running Linux 64 bit OS. The same happens with Spinhenge@home. Other projects binaries are fine, except Rosetta@home which would crash and corrupt all running Rosetta if more than about 1/3 of teh cpu's are trying to run at the same time rosetta... Coming back to Einstein@home binaries (and Spinhenge@home) they run for even less than a second before failing. Any ideea what I must do? Thank you. Ionel ____________
ID: 111036 \| Reply Quote

Richard Haselgrove Send message Joined: Dec 10 05 Posts: 1123 Credit: 22,345,687 RAC: 18,258	Message 111038 - Posted 10 Mar 2011 14:40:17 UTC - in response to Message 111036.
Post some links to failed tasks so that the error messages can be inspected? Failing that, my guess is that you still have to install the 32-bit compatibility libraries - Einstein Linux apps are all 32-bit, and that's a mistake which many users before you have made.
ID: 111038 \| Reply Quote

Bernd Machenschalk Forum moderator Project administrator Project developer Send message Joined: Oct 15 04 Posts: 2975 Credit: 76,670,806 RAC: 33,464	Message 111039 - Posted 10 Mar 2011 16:32:05 UTC Last modified: 10 Mar 2011 16:36:04 UTC
Well, in stderr of this tasks I read: error while loading shared libraries: libdl.so.2: cannot open shared object file: No such file or directory From the "Application version" I also see that you are running the "SSE2" version of the App, which is 32Bit, though I'm not sure why. Could be that you're running a 32 Bit BOINC Core Client or the Client falsely assumes that you have installed the 32 Bit compatibility libraries. I'd try that first, i.e. install the compatibility libraries for running 32 Bit Applications. My guess is that the libdl.so.2 on your system is 64 Bit only and can't be linked to the 32 Bit App. BM
ID: 111039 \| Reply Quote

Mugurel Send message Joined: Jan 20 06 Posts: 3 Credit: 17,799,195 RAC: 10,407	Message 111045 - Posted 10 Mar 2011 22:29:31 UTC
I can only find stuff like this in the log file: 10-Mar-2011 07:18:34 [Einstein@Home] Starting h1_1485.00_S5R4__1218_S5GC1HFa_1 10-Mar-2011 07:18:35 [Einstein@Home] Starting task h1_1485.00_S5R4__1218_S5GC1HFa_1 using einstein_S5GC1HF version 107 10-Mar-2011 07:18:35 [Einstein@Home] Computation for task h1_1485.00_S5R4__1218_S5GC1HFa_1 finished 10-Mar-2011 07:18:35 [Einstein@Home] Output file h1_1485.00_S5R4__1218_S5GC1HFa_1_0 for task h1_1485.00_S5R4__1218_S5GC1HFa_1 absent 10-Mar-2011 07:18:35 [Einstein@Home] Starting h1_1484.95_S5R4__1163_S5GC1HFa_2 10-Mar-2011 07:18:36 [Einstein@Home] Starting task h1_1484.95_S5R4__1163_S5GC1HFa_2 using einstein_S5GC1HF version 107 10-Mar-2011 07:18:36 [Einstein@Home] Computation for task h1_1484.95_S5R4__1163_S5GC1HFa_2 finished 10-Mar-2011 07:18:36 [Einstein@Home] Output file h1_1484.95_S5R4__1163_S5GC1HFa_2_0 for task h1_1484.95_S5R4__1163_S5GC1HFa_2 absent 10-Mar-2011 07:18:36 [Einstein@Home] Starting h1_1484.95_S5R4__1153_S5GC1HFa_0 10-Mar-2011 07:18:37 [Einstein@Home] Starting task h1_1484.95_S5R4__1153_S5GC1HFa_0 using einstein_S5GC1HF version 107 10-Mar-2011 07:18:38 [Einstein@Home] Computation for task h1_1484.95_S5R4__1153_S5GC1HFa_0 finished 10-Mar-2011 07:18:38 [Einstein@Home] Output file h1_1484.95_S5R4__1153_S5GC1HFa_0_0 for task h1_1484.95_S5R4__1153_S5GC1HFa_0 absent 10-Mar-2011 07:18:38 [Einstein@Home] Starting h1_1484.95_S5R4__1152_S5GC1HFa_0 10-Mar-2011 07:18:39 [Einstein@Home] Starting task h1_1484.95_S5R4__1152_S5GC1HFa_0 using einstein_S5GC1HF version 107 10-Mar-2011 07:18:39 [Einstein@Home] Computation for task h1_1484.95_S5R4__1152_S5GC1HFa_0 finished 10-Mar-2011 07:18:39 [Einstein@Home] Output file h1_1484.95_S5R4__1152_S5GC1HFa_0_0 for task h1_1484.95_S5R4__1152_S5GC1HFa_0 absent I run this boinc version: boinc_6.10.58_x86_64-pc-linux-gnu . ldd says: ldd boinc_6.10.58_x86_64-pc-linux-gnu linux-vdso.so.1 => (0x00007fff6d7ff000) libdl.so.2 => /lib/libdl.so.2 (0x00007f1b14928000) libnsl.so.1 => /lib/libnsl.so.1 (0x00007f1b1470e000) libz.so.1 => /lib/libz.so.1 (0x00007f1b144f6000) libpthread.so.0 => /lib/libpthread.so.0 (0x00007f1b142d9000) libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x00007f1b13fc5000) libm.so.6 => /lib/libm.so.6 (0x00007f1b13d41000) libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x00007f1b13b2a000) libc.so.6 => /lib/libc.so.6 (0x00007f1b137a7000) /lib64/ld-linux-x86-64.so.2 (0x00007f1b14b3c000) Going into projects/einstein.phys.uwm.edu and running this: ldd linux-gnu \| grep not \| sort -u gives me: libGL.so.1 => not found libdl.so.2 => not found libm.so.6 => not found I can't figure out which packages I must install for those libraries... I will try few. In the mean time, why ldd on the boinc executable says libdl.so.2 => /lib/libdl.so.2 (0x00007f024a895000) while ldd on one einstein executable says libdl.so.2 => not found ? I guess that is not my fault. Ionel ____________
ID: 111045 \| Reply Quote

Bernd Machenschalk Forum moderator Project administrator Project developer Send message Joined: Oct 15 04 Posts: 2975 Credit: 76,670,806 RAC: 33,464	Message 111047 - Posted 10 Mar 2011 23:30:22 UTC - in response to Message 111045. Last modified: 10 Mar 2011 23:31:25 UTC
See section "64 Bit Considerations" here. The loader of your 64 bit system can link the 64 bit version of libm.so.6 and libdl.so.2 to the 64 Bit BOINC Client, but not to the 32 Bit Applications. The trouble is that the BOINC Client reports that your system is capable of running 32 Bit Apps, which it actually isn't. Possibly there is some way you could tell the Client to omit that (i.e. disable the i686-pc-linux-gnu platform), but I don't know. There is code in BOINC to actually check for the 32 Bit libraries being present before reporting the 32 Bit platform, but this is pretty new. I don't think it's in the 6.10.58 Client, I guess it's only in some recent versions of 6.12. Einstein@home has a Linux 64 Bit application that should work on your system, but being slightly slower than the 32 Bit SSE2 App you'll get the latter if your client reports it can run 32 Bit Apps. BM
ID: 111047 \| Reply Quote

Mugurel Send message Joined: Jan 20 06 Posts: 3 Credit: 17,799,195 RAC: 10,407	Message 111055 - Posted 11 Mar 2011 8:37:22 UTC
Hi, I solved the problem using brute-force! :-) I simply copied /lib32/libm.so.6 /lib32/libdl.so.2 and the two they actually link to: /lib32/libdl-2.11.1.so /lib32/libm-2.11.1.so from a computer which has the same architecture and were it works (e.g. the master node of those nodes). All is in order now. Some nodes must wait a while now because they reach a limit of 16 tasks per day. I need now to solve the same problem for Spinhenge@home. The above trick did not work. More libraries are missing... Thank you for explaining. Ionel P.S. Totally unrelated, Bernd, I am considering for a while now to compile Einstein@home for IRIX. Is there a place I can get some help installing the missing libraries on my IRIX so I can complete the compilation? I will share the executable with other IRIX users when/if I succeed. ____________
ID: 111055 \| Reply Quote

Bernd Machenschalk Forum moderator Project administrator Project developer Send message Joined: Oct 15 04 Posts: 2975 Credit: 76,670,806 RAC: 33,464	Message 111057 - Posted 11 Mar 2011 14:29:15 UTC - in response to Message 111055. Last modified: 11 Mar 2011 19:03:08 UTC
Totally unrelated, Bernd, I am considering for a while now to compile Einstein@home for IRIX. Is there a place I can get some help installing the missing libraries on my IRIX so I can complete the compilation? I will share the executable with other IRIX users when/if I succeed. I'd try: install at least autotools, some gcc and wget (might need more) from SGI freeware Download and unpack this archive. Run the file compile.sh from this package. It should download and compile additional source tarballs and install this software in /usr/local. This includes a recent version of gcc, newer autotools and git. See the license page for the actual source code of the GW application, including a build script. Feel free to PM me if you have further questions or ask these in the forums. But please understand that support for self built applications is not our highest priority. BM
ID: 111057 \| Reply Quote

Author

Message

Mugurel
Send message
Joined: Jan 20 06
Posts: 3
Credit: 17,799,195
RAC: 10,407

Hi,

Einstein@home binaries would fail to run on my new 48cores nodes running Linux 64 bit OS. The same happens with Spinhenge@home. Other projects binaries are fine, except Rosetta@home which would crash and corrupt all running Rosetta if more than about 1/3 of teh cpu's are trying to run at the same time rosetta...

Coming back to Einstein@home binaries (and Spinhenge@home) they run for even less than a second before failing.

Any ideea what I must do?

Thank you.

Ionel
____________

ID: 111036 |
Reply
Quote

Richard Haselgrove
Send message
Joined: Dec 10 05
Posts: 1123
Credit: 22,345,687
RAC: 18,258

Post some links to failed tasks so that the error messages can be inspected?

Failing that, my guess is that you still have to install the 32-bit compatibility libraries - Einstein Linux apps are all 32-bit, and that's a mistake which many users before you have made.

ID: 111038 |
Reply
Quote

Bernd Machenschalk
Forum moderator
Project administrator
Project developer
Avatar

Send message
Joined: Oct 15 04
Posts: 2975
Credit: 76,670,806
RAC: 33,464

Well, in stderr of this tasks I read:

error while loading shared libraries: libdl.so.2: cannot open shared object file: No such file or directory

From the "Application version" I also see that you are running the "SSE2" version of the App, which is 32Bit, though I'm not sure why. Could be that you're running a 32 Bit BOINC Core Client or the Client falsely assumes that you have installed the 32 Bit compatibility libraries. I'd try that first, i.e. install the compatibility libraries for running 32 Bit Applications. My guess is that the libdl.so.2 on your system is 64 Bit only and can't be linked to the 32 Bit App.

BM

ID: 111039 |
Reply
Quote

Mugurel
Send message
Joined: Jan 20 06
Posts: 3
Credit: 17,799,195
RAC: 10,407

I can only find stuff like this in the log file:

10-Mar-2011 07:18:34 [Einstein@Home] Starting h1_1485.00_S5R4__1218_S5GC1HFa_1
10-Mar-2011 07:18:35 [Einstein@Home] Starting task h1_1485.00_S5R4__1218_S5GC1HFa_1 using einstein_S5GC1HF version 107
10-Mar-2011 07:18:35 [Einstein@Home] Computation for task h1_1485.00_S5R4__1218_S5GC1HFa_1 finished
10-Mar-2011 07:18:35 [Einstein@Home] Output file h1_1485.00_S5R4__1218_S5GC1HFa_1_0 for task h1_1485.00_S5R4__1218_S5GC1HFa_1 absent
10-Mar-2011 07:18:35 [Einstein@Home] Starting h1_1484.95_S5R4__1163_S5GC1HFa_2
10-Mar-2011 07:18:36 [Einstein@Home] Starting task h1_1484.95_S5R4__1163_S5GC1HFa_2 using einstein_S5GC1HF version 107
10-Mar-2011 07:18:36 [Einstein@Home] Computation for task h1_1484.95_S5R4__1163_S5GC1HFa_2 finished
10-Mar-2011 07:18:36 [Einstein@Home] Output file h1_1484.95_S5R4__1163_S5GC1HFa_2_0 for task h1_1484.95_S5R4__1163_S5GC1HFa_2 absent
10-Mar-2011 07:18:36 [Einstein@Home] Starting h1_1484.95_S5R4__1153_S5GC1HFa_0
10-Mar-2011 07:18:37 [Einstein@Home] Starting task h1_1484.95_S5R4__1153_S5GC1HFa_0 using einstein_S5GC1HF version 107
10-Mar-2011 07:18:38 [Einstein@Home] Computation for task h1_1484.95_S5R4__1153_S5GC1HFa_0 finished
10-Mar-2011 07:18:38 [Einstein@Home] Output file h1_1484.95_S5R4__1153_S5GC1HFa_0_0 for task h1_1484.95_S5R4__1153_S5GC1HFa_0 absent
10-Mar-2011 07:18:38 [Einstein@Home] Starting h1_1484.95_S5R4__1152_S5GC1HFa_0
10-Mar-2011 07:18:39 [Einstein@Home] Starting task h1_1484.95_S5R4__1152_S5GC1HFa_0 using einstein_S5GC1HF version 107
10-Mar-2011 07:18:39 [Einstein@Home] Computation for task h1_1484.95_S5R4__1152_S5GC1HFa_0 finished
10-Mar-2011 07:18:39 [Einstein@Home] Output file h1_1484.95_S5R4__1152_S5GC1HFa_0_0 for task h1_1484.95_S5R4__1152_S5GC1HFa_0 absent

I run this boinc version:

boinc_6.10.58_x86_64-pc-linux-gnu

.
ldd says:

ldd boinc_6.10.58_x86_64-pc-linux-gnu
linux-vdso.so.1 => (0x00007fff6d7ff000)
libdl.so.2 => /lib/libdl.so.2 (0x00007f1b14928000)
libnsl.so.1 => /lib/libnsl.so.1 (0x00007f1b1470e000)
libz.so.1 => /lib/libz.so.1 (0x00007f1b144f6000)
libpthread.so.0 => /lib/libpthread.so.0 (0x00007f1b142d9000)
libstdc++.so.6 => /usr/lib/libstdc++.so.6 (0x00007f1b13fc5000)
libm.so.6 => /lib/libm.so.6 (0x00007f1b13d41000)
libgcc_s.so.1 => /lib/libgcc_s.so.1 (0x00007f1b13b2a000)
libc.so.6 => /lib/libc.so.6 (0x00007f1b137a7000)
/lib64/ld-linux-x86-64.so.2 (0x00007f1b14b3c000)

Going into projects/einstein.phys.uwm.edu and running this:

ldd *linux-gnu* | grep not | sort -u

gives me:

libGL.so.1 => not found
libdl.so.2 => not found
libm.so.6 => not found

I can't figure out which packages I must install for those libraries... I will try few.

In the mean time, why ldd on the boinc executable says

libdl.so.2 => /lib/libdl.so.2 (0x00007f024a895000)

while ldd on one einstein executable says

libdl.so.2 => not found

?
I guess that is not my fault.

Ionel
____________

ID: 111045 |
Reply
Quote

Bernd Machenschalk
Forum moderator
Project administrator
Project developer
Avatar

Send message
Joined: Oct 15 04
Posts: 2975
Credit: 76,670,806
RAC: 33,464

See section "64 Bit Considerations" here.

The loader of your 64 bit system can link the 64 bit version of libm.so.6 and libdl.so.2 to the 64 Bit BOINC Client, but not to the 32 Bit Applications.

The trouble is that the BOINC Client reports that your system is capable of running 32 Bit Apps, which it actually isn't.

Possibly there is some way you could tell the Client to omit that (i.e. disable the i686-pc-linux-gnu platform), but I don't know.

There is code in BOINC to actually check for the 32 Bit libraries being present before reporting the 32 Bit platform, but this is pretty new. I don't think it's in the 6.10.58 Client, I guess it's only in some recent versions of 6.12.

Einstein@home has a Linux 64 Bit application that should work on your system, but being slightly slower than the 32 Bit SSE2 App you'll get the latter if your client reports it can run 32 Bit Apps.

BM

ID: 111047 |
Reply
Quote

Mugurel
Send message
Joined: Jan 20 06
Posts: 3
Credit: 17,799,195
RAC: 10,407

Hi,

I solved the problem using brute-force! :-)

I simply copied

/lib32/libm.so.6
/lib32/libdl.so.2

and the two they actually link to:

/lib32/libdl-2.11.1.so
/lib32/libm-2.11.1.so

from a computer which has the same architecture and were it works (e.g. the master node of those nodes).
All is in order now. Some nodes must wait a while now because they reach a limit of 16 tasks per day.

I need now to solve the same problem for Spinhenge@home. The above trick did not work. More libraries are missing...

Thank you for explaining.

Ionel

P.S. Totally unrelated, Bernd, I am considering for a while now to compile Einstein@home for IRIX. Is there a place I can get some help installing the missing libraries on my IRIX so I can complete the compilation? I will share the executable with other IRIX users when/if I succeed.
____________

ID: 111055 |
Reply
Quote

Bernd Machenschalk
Forum moderator
Project administrator
Project developer
Avatar

Send message
Joined: Oct 15 04
Posts: 2975
Credit: 76,670,806
RAC: 33,464

Totally unrelated, Bernd, I am considering for a while now to compile Einstein@home for IRIX. Is there a place I can get some help installing the missing libraries on my IRIX so I can complete the compilation? I will share the executable with other IRIX users when/if I succeed.

I'd try:

install at least autotools, some gcc and wget (might need more) from SGI freeware
Download and unpack this archive. Run the file compile.sh from this package. It should download and compile additional source tarballs and install this software in /usr/local. This includes a recent version of gcc, newer autotools and git.
See the license page for the actual source code of the GW application, including a build script.

Feel free to PM me if you have further questions or ask these in the forums. But please understand that support for self built applications is not our highest priority.

BM

ID: 111057 |
Reply
Quote