Posts by Hermes

1) Message boards : SZTAKI Desktop Grid : Validation Problems - Windows vs. Linux (Message 6575)
Posted 3743 days ago by Hermes
WU 571699
2) Message boards : SZTAKI Desktop Grid : Checkpointing/Restart issue? (Message 3782)
Posted 4124 days ago by Hermes
Is there a problem with restarting from a checkpoint?

I noticed that Result 12833 exited immediatly after restarting:
2006-08-10 03:18:53 [SZTAKI Desktop Grid] Restarting task 700907a0-4bc7-46cd-879e-e5dc92c4a78c_518673c3-cdad-4f4f-b311-a4470ea49f7b_265_3 using search version 200
2006-08-10 03:18:54 [---] Rescheduling CPU: application exited
2006-08-10 03:18:54 [SZTAKI Desktop Grid] Computation for task 700907a0-4bc7-46cd-879e-e5dc92c4a78c_518673c3-cdad-4f4f-b311-a4470ea49f7b_265_3 finished


The other 2 successful results from that WU seem to have restarted also, and the claimed credits differ significantly.
3) Message boards : SZTAKI Desktop Grid : New applications 1.10 online - Feedback (Message 2718)
Posted 4207 days ago by Hermes
The linux version still doesn\'t save the checkpoint CPU time.
4) Message boards : SZTAKI Desktop Grid : New applications 1.06 and 1.07 online - Feedback (Message 2453)
Posted 4240 days ago by Hermes
Linux App 1.07 under Boinc CC 5.2.13
Checkpointing does not work: the checkpoint CPU time is not set, so a continued WU is started at the checkpointed tiem but with 0 CPU time.

Without a restart form a checkpoint the result of the Linux application is marked invalid when compared with results from the Windows application. See WU 510487
5) Message boards : SZTAKI Desktop Grid : Ad@m\'s BLOG (page 2) (Message 1348)
Posted 4361 days ago by Hermes
It seems that the new WU layout works and the system is stable, so there is nothing that could stop us now from reaching beyond 500 GFlops.


That is: nothing but the daily quota bug.
After almost 3 days of waiting I finally got a new WU. My computer immediatly began the immensly time consuming task of calculating the problem. It was a hard fight. And for al long time it was unclear, who would be the winner. But after 3 (in words: three) seconds (!) the solution was found. Now I will have to wait another 3 days before I get the next WU.

TBC
6) Message boards : Észrevételek, tapasztalatok : Dramatic drop in GFlops/Sec (Message 1330)
Posted 4362 days ago by Hermes
I\'ve noticed that the Gflops/Sec has dropped dramatically in the last few hours. It had been going up steadly after the upgrade. Any idea why this current drop.

Perhaps because of the Daily quota problem?
7) Message boards : Észrevételek, tapasztalatok : 0 credits because of too many total results (Message 705)
Posted 4483 days ago by Hermes
I have 2 more with credit only granted to 1 of 3 results. How can 1 out of 3 have a majority to form the quorum?
<a href="http://szdg.lpds.sztaki.hu/szdg/workunit.php?wuid=1594">WU 1594</a>
<a href="http://szdg.lpds.sztaki.hu/szdg/workunit.php?wuid=1628">WU 1628</a>

And next time you send a mass EMail to inform us: Don't put in all of the recipients addresses in the TO-Field for everyone to see. That's what the BCC-Field is for.
8) Message boards : Észrevételek, tapasztalatok : 0 credit for a correct WU (Message 598)
Posted 4491 days ago by Hermes
This bug affects all other Boinc projects, not just Sztaki DG.

I think the Boinc Developers shoould take a close look at there code and find out, why BURP can interfere with other projects. Such things must not be possible. This time it may be sloppy programming, next time someone might use these loopholes to deliberatly sabotage another project.
9) Message boards : Észrevételek, tapasztalatok : PLEASE EXPLAIN THIS WORK UNIT (Message 591)
Posted 4491 days ago by Hermes
Predictor@home has set "Homogeneous Redundancy" in their scheduler so it sends the results for one work unit to just one platform/CPU-vendor combination. Einstein@home has relaxed the restrictions on the validator to accept results that have slight differences. Altough it took some time for them to find the right amount of "difference" to allow.
Most of the other projects will probably have implemented one or the other solution by now.

I just wonder, why every projekt has to make the same mistakes all over again? Is there no "Howto make a properly running Boinc Projekt", that warns of the common problems? Like having an underpowered database, setting a too tight deadline and that the different compilers from different OSes produce diffenrent code that will almost certainly lead to different results. (At least when using floating point math)


Home | My Account | Message Boards


Copyright © 2017 SZTAKI Desktop Grid