Testing of BOINC version of theSkyNet

[Mike Mitchell]

I just joined the project, and added the team, got one work unit. I'll run that and see what it's like before making plans. I'm only the 300^th project member.

Hmm, I have an idea for the next AA. 

[Dataman]

I got 4 but big problems ... NO CHECKPOINTS????? I did my normal weekly PM on that machine and restarted at 0 time. No can do on 14-15 hour jobs. Any one getting checkpoints?  :hbang:

[JugNut]

Thanks for the update lawry.  I've had similar problems as well.  As I may have mentioned before I have two i7's one is 100% stable & one is at best 85% stable & much to much a risk on 12hr+ non check-pointing WU's.     I've stopped crunching on it & have other work that needs crunching also.  So while i'll keep a few core's with the light on, most of my effort will have to resume in a few days time. (hopefully when the check-pointing issue is fixed)

Cheers ..JN..

[Dataman]

Yep, definately not checkpointing Windows. Have set it to no new tasks until I see something in the form.

HOT here and not much crunching going on. 106F

:aus1: :US

[Dingo]

Sorry guys I thought it was fixed for all Operating Systems.   

[Dataman]

No harm no foul, Dingo. I'll keep watching. 

[kashi]

I had one task restart from a checkpoint successfully. "Leave applications in memory while suspended" was turned off when my preferences accidentally got reset. Have not tried exiting BOINC or rebooting though. Restart at fit 24 is shown in Stderr output:
"04:33:47 (4296): wrapper: running fit_sed (24 filters.dat observations.dat)
wrapper: starting
05:09:47 (2548): wrapper: running fit_sed (24 filters.dat observations.dat)"

After the checkpoint restart, Elapsed time as shown in BOINC Manager reset to zero or one  "fit" unit's runtime but percentage completed was correct. This mucked up "Remaining (estimated)" time too. This is all to do with how the wrapper mechanism interacts with BOINC.

When the task's Elapsed time reset I was concerned that the checkpoint hadn't worked but task completed without any trouble. 

All the tasks I have looked at are successfully writing checkpoint files as seen in wrapper_checkpoint.txt file. Because this application uses a wrapper, "CPU time at last checkpoint" does not show anything in BOINC Manager task properties. It will only show something there (one fit duration) if the application has actually restarted from a checkpoint.

At least that's how a restart was shown on this Windows 7 computer. Other operating systems may behave and/or report things differently.

[kashi]

What I meant to imply by my previous post is that due to the way it is reporting some people may think that checkpointing is not working when it is. For example, Dataman's 4 aborted tasks show 3 fits successfully completed and 2 restarts on fit 4, although what was shown to him by BOINC Manager made him think those tasks had restarted from the beginning.

Wait for one fit duration (often about 2% of total task time) then have a look in your stderr.txt file for that task in your BOINC slots folder if you think checkpointing hasn't worked.

[JugNut]

Does anyone use BoincTasks? It track's checkpoints in real time & also CPU useage. (may have to be activated in settings?)

It paints a strange picture, checkpoints are always in the red (overtime) & CPU usage shows at between 2 & 15% also in the red? What the Fred does that mean?

Anyway it's all good Dingo just minor teething probs, we'll crunch up a storm when all the small bugs are fixed..

Cheers ..JN..

[kashi]

If BoincTasks is based on the same information as BOINC Manager, it will have the same reporting limitations with these POGS wrapper tasks. In other words it will not display checkpoints or CPU time correctly, the same as BOINC Manager. With these wrapper tasks on Windows the CPU time resets at the end of each fit, that's why most completed Windows tasks show zero CPU time reported on the POGS website after the concatenation processing. For the same reason, during the task processing the BOINC Manager CPU time in Properties shows the time since the last fit was started, which is also the time since the last checkpoint.

[Dingo]

The next AA starts on the 1st September.  I have sent a message to Kevin at theSkyNet to see if he thinks it can handle an AA.

Hi Kevin,

Doing a great job with the project, only complaints at the present from my members is that check pointing is not working on Windows PC's.

Four times a year the Team has an Aussie Assault (AA) where the team picks a project that we get as many Team members as possible to work on at one time. This boosts the Team together and we usually pick a project that we need a boost in the standings on.

The next AA is due in September and I am thinking that I will propose this project as it is the first Australian Project. The team has over 4000 members but only about a 1000 usually join but because the project is Australian there may be more.

Do you think your project can handle the work load and there will be enough work available between 01 - 14 September 2012.

Cheers
Dingo

I just got his reply which looks like it may be ready he does't know, especially if he needs to write his own wrapper.

I thing I'll have to ditch the wrapper and write my own :-(

I've just got 216 new galaxies - so maybe....

We can test it over the next two weeks

I will put it up as an option and if it is not ready we can do the AA on the alternate.

[JugNut]

Excellent news Dingo.  I'll keep my fingers crossed. 

Like Lawry said and previously by kashi the check-pointing does work for me also.    It just doesn't appear to be working as kashi mentioned earlier.  As usual Kashi's right on the money.

Anyway things are looking up on the theSkyNet front, I wish the WU's were a bit smaller & the stats a bit more regular, but you can't have everything I guess.   And an AA would be a great way to help spread the word & help bolster our score.  Sooo...

Bring on the AA..

Cheers to all ..JN..

[Dataman]

Ok mates, thanks for the information. I'll try to grab another couple of wu's and watch them. Still little crunching going on here ... hot Hot HOT! (104F)

[Dataman]

POGS hates me.  :hbang: I started 8 on a i7 920 and they are running at 22-27% CPU. I set it back to no new tasks but will try to let them complete but it is going to take 4x the time. Maybe they will use more CPU later.  :cry2:

[kashi]

Is it showing only 22-27% total CPU Usage in Windows Task Manager for 8 tasks combined? Very strange, I'm running 7 cores and total CPU Usage is 87% as expected. Maybe running all 8 cores is bogging it down, especially if you are running a GPU project at the same time. Perhaps you could try 7 cores (87.5% of processors) and see if it improves.

You're not using BoincTasks to monitor CPU Usage are you? That will give incorrect CPU usage readings on this wrapper project in Windows as reported by JugNut here and Saenger on the project forum.