S5R2 phones home on compute error?
24 Apr 2007 17:09:49 UTC
Topic 192624
(moderation:
My first S5R2 result errored out.
result 83459803
I noticed it had turned yellow (failure to progress at normal rate) in Boincview as viewed on my monitoring computer. On checking the computer running the result, I found that ZoneAlarm was alerting me that the Einstein science application was "trying to access the internet". When I clicked the authorization button, it promptly moved from running (but accumulating no CPU) state to errored out.
Perhaps the previous science aps did something similar, but I don't recall it. Possibly I authorized them a while back, or possibly I've had no compute errors on Einstein in the life of this machine.
As to the error itself, at the top of stderr it says:
Quote:
exit code -1073741819 (0xc0000005)
and just before the dubugger section:
Quote:
Unhandled Exception Detected...
- Unhandled Exception Record -
Reason: Access Violation (0xc0000005) at address 0x0044AE0C read attempt to address 0x00000899
These seem common symptoms.
This machine is a Core 2 Duo E6600 running at an overclock frequency/voltage combination (3.006 Ghz, 1.34375V)that is just barely adequate to allow KWSN SETI to run for a couple of days without error, so the possibility that this is just a speed problem is open for my machine.
S5R2 phones home on compute error?
)
Hmmmm, that does sound strange since AFAIK all the comm traffic to the project is supposed to be handled by the CC.
The only thing I can think of at the moment is ZA was actually complaining about an application hijack attempt or perhaps a file signature check, and the wording of the alarm message is misleading.
I guess the easiest way to test that is to just clear all the BOINC related items from the trusted apps list and then work through the ZA prompts again.
Alinator
RE: I guess the easiest
)
Your comment prompted me to go in and look at program-level permissions. That clarified that I had previously set the same permissions for S5RI_4.24 as I have now set for S5R2_4.13.
This makes it seem likely to me that the two versions are doing the same thing in this respect.
OK, then it does sounds like
)
OK, then it does sounds like this was just an app control issue for ZA and most likely the wording of the message could be more specific/accurate. I haven't run ZA for a while, but I remember when I did you had to drill down a bit to get the full story on what warnings really meant sometimes.
The good news is at least you know ZA did what it was suppose to. ;-)
Alinator
I notice the line Symbol
)
I notice the line
Symbol Search Path: C:\\Program Files\\BOINC\\slots\\0;C:\\Program Files\\BOINC\\projects\\einstein.phys.uwm.edu;srv*C:\\DOCUME~1\\Peter\\LOCALS~1\\Temp\\symbols*http://msdl.microsoft.com/download/symbols
in the crash output - so quite possibly the science application itself does contact the internet (to get debug information) in the event of a crash, even though all routine project communication is handled by the BOINC CC.
RE: This machine is a Core
)
As of this writing, all three of the first three S5R2 results attempted on this C2D overclocked machine have errored out within the first hour of a predicted 7 hour run time.
[/url=http://einsteinathome.org/task/83459803]result 83459803[/url]
[/url=http://einsteinathome.org/task/83497207]result 83497207[/url]
[/url=http://einsteinathome.org/task/83502778]result 83502778[/url]
Pursuing the possibility that the real problem here is one of speed, I've raised the CPU voltage by five whole increments (i.e. from 1.34375 to 1.375V). If I don't have errors in the next day, I'll try lowering the voltage again. If I can turn this thing off and on with voltage (meaning speed), perhaps others will be willing to entertain the possibility that some/much/all of the problems seen on other machines may be speed related, and consider checking by slowing clock or raising voltage appreciably.
On this same machine, the previous Einstein ran for a couple of days flawlessly at 3.006 GHz, 1.31875V. It was occasional SETI errors (always access violations) that forced me up increment by increment until finally 1.34375 ran dozens of results over a couple days flawlessly. Perhaps this Einstein ap happens to create a condition that is yet a bit more speed-challenged. No accusation against either Intel (which only claimed my part would do 2.4 GHz at 1.3275) or the Einstein developers (who can't possibly know what legal code may tickle a little-used slow path) applies if this turns out to be true.
RE: If I can turn this
)
I've done much further work, which I think corroborates that my host, at least, had a fixable speed problem in running S5R2. For much more detail, see a thread a thread I've started in the Problems and Bug Reports forum.