Computation Error on 32-bit Linux


Message boards : Number crunching : Computation Error on 32-bit Linux

Message board moderation

To post messages, you must log in.
AuthorMessage
Profile [AF>Dell>UniversdeDell.com] At...
Avatar

Send message
Joined: 5 May 20
Posts: 33
Credit: 42,115,149
RAC: 1
Message 7066 - Posted: 18 Sep 2020, 17:54:49 UTC
If I'm not wrong, Ubuntu 18.04 use GLIBC version 2.27, this would explain your error.
ID: 7066 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bigmike206

Send message
Joined: 26 Oct 20
Posts: 5
Credit: 348,505
RAC: 85
Message 7094 - Posted: 15 Nov 2020, 14:52:00 UTC - in response to Message 7066.  

Last modified: 15 Nov 2020, 15:06:33 UTC
If I'm not wrong, Ubuntu 18.04 use GLIBC version 2.27, this would explain your error.

I'm experiencing the same problem. You suggest it may be due to the GLIBC version. If so, what version of GLIBC is required?

It seems strange, as all of my machines have the same GLIBC (v2.23): but the 64-bit machines process A@H work flawlessly, while all the 32-bit boxes error out every A@H task in about 1 second.

Follow-up question: is it sufficient to simply upgrade my GLIBC package, or would I also need to re-compile BOINC afterwards?
ID: 7094 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile [AF>Dell>UniversdeDell.com] At...
Avatar

Send message
Joined: 5 May 20
Posts: 33
Credit: 42,115,149
RAC: 1
Message 7097 - Posted: 17 Nov 2020, 21:22:11 UTC - in response to Message 7094.  
Unless this is not a critical machine (and even if not critical by the way) NEVER EVER try to update/upgrade such a package, you'll, most of the time, broke your whole system.

On the initial post, we can see "/lib/i386-linux-gnu/libc.so.6: version `GLIBC_2.28' not found" so it didn't found the version 2.28, at least, this is why I said that, there could be some different requirements between x86 and x64 app version too, I don't know.

I'm only using Debian, so can't tell for sure what's wrong on your end.
ID: 7097 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bigmike206

Send message
Joined: 26 Oct 20
Posts: 5
Credit: 348,505
RAC: 85
Message 7098 - Posted: 18 Nov 2020, 1:41:23 UTC - in response to Message 7097.  

Last modified: 18 Nov 2020, 1:53:55 UTC
On the initial post, we can see "/lib/i386-linux-gnu/libc.so.6: version `GLIBC_2.28' not found" so it didn't found the version 2.28, at least, this is why I said that, ...

Gotcha. Thanks for the response. (Sorry I missed that in the OP's error message.)

Unless this is not a critical machine (and even if not critical by the way) NEVER EVER try to update/upgrade such a package, you'll, most of the time, broke your whole system.

Since my last post, I've gone ahead and tried updating GLIBC to version 2.30 on a couple of my 32-bit boxes and that has seemed to work like a charm: they're computing now without error. (And I didn't have to recompile BOINC.)

No problems with any of my other software, so far. Which isn't very much, honestly. These are old stripped down rigs, running headless, with a minimal O.S.--just a kernel and enough packages installed for basic CLI system admin, networking, and to compile BOINC client from source. I.e., they'll do nothing but crunch until they've expired! ;)
ID: 7098 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bigmike206

Send message
Joined: 26 Oct 20
Posts: 5
Credit: 348,505
RAC: 85
Message 7099 - Posted: 19 Nov 2020, 19:59:44 UTC - in response to Message 7098.  

Last modified: 19 Nov 2020, 20:01:16 UTC
Since my last post, I've gone ahead and tried updating GLIBC to version 2.30 on a couple of my 32-bit boxes and that has seemed to work like a charm: they're computing now without error.

UPDATE: The "standard" version (102.13) tasks are "computing now without error." However, the "sse2" versions are still failing in about 1 second, with a different message:

<message>
process exited with code 193 (0xc1, -63)</message>
<stderr_txt>
SIGILL: illegal instruction
Stack trace (5 frames):
../../projects/asteroidsathome.net_boinc/period_search_10213_i686-pc-linux-gnu__sse2_linux(+0x321d6)[0x4771d6]
linux-gate.so.1(__kernel_sigreturn+0x0)[0xb778db40]
../../projects/asteroidsathome.net_boinc/period_search_10213_i686-pc-linux-gnu__sse2_linux(+0x141c5)[0x4591c5]
/lib/libc.so.6(__libc_start_main+0x105)[0xb7451e75]
../../projects/asteroidsathome.net_boinc/period_search_10213_i686-pc-linux-gnu__sse2_linux(+0xbe51)[0x450e51]

Exiting...


(And I didn't have to recompile BOINC.)

In light of the sse2 failures, I've gone ahead and recompiled my BOINC client against my updated GLIBC. Will report back once I get some new sse2 tasks...
ID: 7099 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
bigmike206

Send message
Joined: 26 Oct 20
Posts: 5
Credit: 348,505
RAC: 85
Message 7100 - Posted: 21 Nov 2020, 16:58:31 UTC - in response to Message 7099.  

Last modified: 21 Nov 2020, 16:59:47 UTC
In light of the sse2 failures, I've gone ahead and recompiled my BOINC client against my updated GLIBC. Will report back once I get some new sse2 tasks...
So, I recompiled BOINC, but no dice. All sse2 tasks are still failing in about one second. E.g.,

http://asteroidsathome.net/boinc/result.php?resultid=327113407

I guess it's 64-bit A@H only for me... My four 32-bit boxes will have to stick to Milkyway, Einstein, and Rosetta.
ID: 7100 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote

Message boards : Number crunching : Computation Error on 32-bit Linux