Since I downloaded the updated HPF2 program on 06/28/2006, I have had 1 error and 4 invalids. Here they are, together with my Computer ID, etc, from the Messages Tab in BOINC Manager.

Starting BOINC client version 5.4.9 for windows_intelx86
Processor: 1 AuthenticAMD AMD Sempron(tm) Processor 3100+
Memory: 895.48 MB physical, 2.12 GB virtual
Disk: 40.33 GB total, 29.13 GB free
Computer ID: 32229;

Work Unit --- Time Sent ------------- Time Returned – CPU Time

Error

za095_ 00852 07/05/2006 21:18:42 07/06/2006 09:23:15 3.68

Invalid

za053_ 00268 07/03/2006 08:35:42 07/03/2006 17:38:16 4.58
za083_ 00265 07/03/2006 01:38:35 07/03/2006 12:49:27 4.48
za082_ 00363 07/02/2006 10:18:15 07/02/2006 19:25:46 4.73
za067_ 00001 07/01/2006 03:25:32 07/01/2006 14:56:31 5.06

I still get abnormal WU terminations and corrupted double linked list messages in the terminal (but less frequently than it used to happen). At least two faults occurred right after the WU started, after 3 or 4 seconds of CPU time, if that can help. za081_00256_2 is one of them (I had to cancel it); I forgot to keep track of the number of the other one (but it's probably za078_0064, which terminated abnormally yesterday, with signal SIGSEGV).
This happens on a P4 2.6 GHz with 512 MB of RAM, running BOINC 5.4.9 under MEPIS/Debian GNU/Linux.

The older computer I put back on the grid in the past few days, which is Athlon 1 GHz with 128 MB of RAM, crunches FAAH WUs correctly, although the BOINC agent complains that the WU needs ~50% more memory than I have. I have seen that HPF2 requires more power or RAM than FAAH, so I guess that's why it has not received HPF2 WUs yet.

Hi Mark099,
Could you give us your Computer ID and the work unit name? Also, which client are you running, what CPU and how much RAM and Virtual Memory do you have?
Lawrence

I have wu currently at 62.5%. I dont know how long exactly it has been stuck, but processing time now totals 21:20 which is excessive for this machine.

I will let it continue, but it is probably wasting a lot of processing time. I think that a self abort feature is important - also maybe an indicator of how long it has been stuck.

Dirk, there is a self aborting, rather a 'skip forward' feature, except in this new science project, not all loops have been identified....i.e. if an attempt is set at 15 minutes, but there are many more attempts than anticipated in a segment of the WU, its going to take much longer. You dont want to skip to the next segment premature, thus its trial and error of where the sensible border lies, to let it rest.

An attempt in representing this in numbers: 1 HPP2 WU is 25 segments:, thus 4% is the factor by which you will see progress on the front GUI page. If progress is measured in the 'i' page as 0.1 percent step, you'd see only 40 times a progression in any segment **. If then a segment is made up of millions *** of possibilities, you'd see this percent progress go forward rarely.

Key is in UD agent to see some progress on the line graphs in the 'i' screen and the tenth of a percent progress bar there, rather than the
full percent progress on the front screen (thus 4 in the example for HPF2). If you dont see any for a long long time, half a day, you can abort by going to taskmanager and kill the UD_7xxxxxxx.exe process. It will send the file back and fetch a new one, and if there is a new version of the science.exe, it will be send with it. I believe to have seen that credit is still given, but the file has to be send back by the agent. Any other way of killing might not do that and simply fetch a new one and WCG recording the cancelled WU as a "no reply". That's waste.

Reports have been requested, to post device number and(approximate) UTC time, local time+zone the WU was downloaded and aborted. Then WCG tech can do detail analysis of the aborted WU, to further tweak the non-convergence loop exits.

In BOINC, its much easier to see. CPU time is clocked in the work tab, not Wallclock. If that is not progressing for a while, it's potentially stuck, but even then i've had ZA's that sat, no CPU time progress for a long while...several hours, NO OTHER PROCESSES HOGGING THE CPU. After that just finishing.

Sofar my log shows 22 HPF2 calculated thru, 2 invalids, 20 valid. I've cancelled 6 or 7 that were in queue but with v 5.06 in BOINC, to force receipt of new HPF2 with version 5.07 (*). Think the equivalent in UD agent is something like 5.05.02 v 5.05.03. You wonder about logs.....yes another very nice feature of BOINC.

* Long story, but any cancelling of 5.06 prior start may have caused excess copies in the 'inconclusive' saga to be crunched. (See Knreed explanations and why up to 12 copies could get send out on a single WU).

** I dont think in the 'i' screen the percent bar skips back to zero for each segment...never observed that. If it does, then obviously, there are 1000x0.1% progression steps visible in a segment. As a suggestion, since i think its up front known by WCG how many segments there are in a HPF1/2 WU, they could put the number in the 'i' page and show e.g. 'segmnt 12 of 25'. That gives together with a 1000x0.1% segment indication on the 'i' progress bar a much greater granularity.

*** One of the explanations of the why it takes much longer was described elsewhere as 'hi-resolution'.....versus HPF1, the nitty gritty detail looked for is much deeper than ever before down to atomic level.

Just my 2 pennies of how i translate it for myself. Anyone is free to skip, amplify or correct. Not interested in comments that start of with 'bad advise....'

time for coffee

ciao