The workunit on my OLPC completed. Unfortunately, the outcome after 873,316 seconds of compute time is "Compute error". What a disappointment. Some guru more knowledgeable than myself will have to tell me what the error means.
The workunit on my OLPC completed. Unfortunately, the outcome after 873,316 seconds of compute time is "Compute error". What a disappointment. Some guru more knowledgeable than myself will have to tell me what the error means.
It failed with a signal 8, which is a floating point exception. That's bad. It means that the hardware detected an inconsistency in it's own calculations. And that in turn could mean that this little box is not really up to the job. That would be disappointing. Or it could be a bug that has nothing to do with this hardware. Bernd could say more than I can.
I now have a little bit better news about my XO. My original test was with "run always" by default, which seems to have caused the laptop to be unresponsive. So I set the host to the 'home' venue, which I configured explicitly to not run when the user is active, and I put CPU throttling to 80%.
* The good news is the CPU throttling works. The einstein_S5R3_4 app uses about 65% of CPU, and never over 80%. The 'uptime' system load is around 0.65. And the XO is completely responsive even as the BOINC application is running.
* The not so good news is that detecting that the user is active does not seem to work. I can use the mouse and keyboard in the Terminal Activity, or use another activity, but this does not cause BOINC to suspend the running application.
The WU it's crunching is very late, because I left the XO off for a long while, so I'm considering aborting it to start another and then letting that one try to run to completion, or to see if it also ends with signal 8.
The WU it's crunching is very late, because I left the XO off for a long while, so I'm considering aborting it to start another and then letting that one try to run to completion, or to see if it also ends with signal 8.
Well, I didn't have to abort the WU, it aborted itself, with signal 8, after only 18,000sec. Then it picked up another one, which also failed with signal 8 after only 5,000sec. So I think there is a clear pattern here.
The WU it's crunching is very late, because I left the XO off for a long while, so I'm considering aborting it to start another and then letting that one try to run to completion, or to see if it also ends with signal 8.
Well, I didn't have to abort the WU, it aborted itself, with signal 8, after only 18,000sec. Then it picked up another one, which also failed with signal 8 after only 5,000sec. So I think there is a clear pattern here.
It has a little green power converter, so it's plugged in to the wall.
Though the batter life seems rather good, so you can unplug it and use it for several hours. I've not tested the limit.
The "wall wart" is thin enough that you could put several of them in a row on a power strip, instead of one of them blocking access to a plug next to it.
The prongs are the same size, so it's not "polarized" - you can plug it in either way and it works.
It says "100-240V 50-60Hz", so I suspect that I could plug it in while in Europe and it would work (with just an adapter to change the pins, not a full transformer), but I have not tried.
Since my last post two new WUs have started and then ended with a client error, so I doubt that it will be possible to ever finish one. I'm thinking maybe I should attach to some other project to see if it's something about the Einstein@Home application, or just a general problem with floating point computing on the XO.
I switched the XO to run SETI@home, and so far it's run 15hrs with no interruptions (except that boinc_client seems to have crashed when I made the initial switch). Fingers crossed.
The first workunit on the XO from SETI@home finished in 31 hours 21 minutes and is "Ready to report". We shall see if it validates. A second WU has begun.
Good news: the second Einstein@Home workunit completed successfully on my XO!
During this workunit, I kept my XO otherwise idle (except occasional ssh into it to check status with boinc_cmd) and powered up continuously with no reboots.
This shows Einstein@Home on the XO is possible in principle and suggests that the failure of my first wu and those that Eric saw were triggered by other activity on the machine.
As I mentioned earlier, I always get a memory allocation failure with yum install, even on a small package, when Einstein@Home is running. This strikes me as odd, because I wouldn't imagine yum as requiring a lot of resources (although I don't really know). It may be a clue to a deeper problem, not sure. The XO machine has 256M of memory, which is twice as much as my Debian machine that has never had a problem with Einstein@Home.
Anyway I've sacrificed 11 days of XO play time for this experiment, which is enough. :) I may keep E@H running while doing other things and will report back if I see any obvious correlation to a failure.
The workunit on my OLPC
)
The workunit on my OLPC completed. Unfortunately, the outcome after 873,316 seconds of compute time is "Compute error". What a disappointment. Some guru more knowledgeable than myself will have to tell me what the error means.
Norm wrote:The workunit on my
)
It failed with a signal 8, which is a floating point exception. That's bad. It means that the hardware detected an inconsistency in it's own calculations. And that in turn could mean that this little box is not really up to the job. That would be disappointing. Or it could be a bug that has nothing to do with this hardware. Bernd could say more than I can.
- Eric Myers
I now have a little bit
)
I now have a little bit better news about my XO. My original test was with "run always" by default, which seems to have caused the laptop to be unresponsive. So I set the host to the 'home' venue, which I configured explicitly to not run when the user is active, and I put CPU throttling to 80%.
* The good news is the CPU throttling works. The einstein_S5R3_4 app uses about 65% of CPU, and never over 80%. The 'uptime' system load is around 0.65. And the XO is completely responsive even as the BOINC application is running.
* The not so good news is that detecting that the user is active does not seem to work. I can use the mouse and keyboard in the Terminal Activity, or use another activity, but this does not cause BOINC to suspend the running application.
The WU it's crunching is very late, because I left the XO off for a long while, so I'm considering aborting it to start another and then letting that one try to run to completion, or to see if it also ends with signal 8.
RE: The WU it's crunching
)
Well, I didn't have to abort the WU, it aborted itself, with signal 8, after only 18,000sec. Then it picked up another one, which also failed with signal 8 after only 5,000sec. So I think there is a clear pattern here.
Here are the results for that host
- Eric Myers
RE: RE: The WU it's
)
Do you run the XO on battery or with a PSU?
Desti wrote:Do you run the XO
)
It has a little green power converter, so it's plugged in to the wall.
Though the batter life seems rather good, so you can unplug it and use it for several hours. I've not tested the limit.
The "wall wart" is thin enough that you could put several of them in a row on a power strip, instead of one of them blocking access to a plug next to it.
The prongs are the same size, so it's not "polarized" - you can plug it in either way and it works.
It says "100-240V 50-60Hz", so I suspect that I could plug it in while in Europe and it would work (with just an adapter to change the pins, not a full transformer), but I have not tried.
Since my last post two new WUs have started and then ended with a client error, so I doubt that it will be possible to ever finish one. I'm thinking maybe I should attach to some other project to see if it's something about the Einstein@Home application, or just a general problem with floating point computing on the XO.
- Eric Myers
I switched the XO to run
)
I switched the XO to run SETI@home, and so far it's run 15hrs with no interruptions (except that boinc_client seems to have crashed when I made the initial switch). Fingers crossed.
- Eric Myers
The first workunit on the XO
)
The first workunit on the XO from SETI@home finished in 31 hours 21 minutes and is "Ready to report". We shall see if it validates. A second WU has begun.
Good news: the second
)
Good news: the second Einstein@Home workunit completed successfully on my XO!
During this workunit, I kept my XO otherwise idle (except occasional ssh into it to check status with boinc_cmd) and powered up continuously with no reboots.
This shows Einstein@Home on the XO is possible in principle and suggests that the failure of my first wu and those that Eric saw were triggered by other activity on the machine.
As I mentioned earlier, I always get a memory allocation failure with yum install, even on a small package, when Einstein@Home is running. This strikes me as odd, because I wouldn't imagine yum as requiring a lot of resources (although I don't really know). It may be a clue to a deeper problem, not sure. The XO machine has 256M of memory, which is twice as much as my Debian machine that has never had a problem with Einstein@Home.
Anyway I've sacrificed 11 days of XO play time for this experiment, which is enough. :) I may keep E@H running while doing other things and will report back if I see any obvious correlation to a failure.
Norm
Congrats Norm! That's pretty
)
Congrats Norm! That's pretty cool.
Kathryn :o)
Einstein@Home Moderator