PDA

View Full Version : FreeHAL: Longer run times?



zombie67
07-31-12, 09:14 PM
Is anyone else getting really long run times now? It used to be ~2-3 hours long, now it is ~10-12 hours long. Anyone have any news?

Duke of Buckingham
08-01-12, 06:32 AM
I have noticed FreeHAL tasks are kind of crazy there are all sorts of tasks. 2-3 hors 6-7 hours or 10-12 hours, I noticed also the tasks are mixed and the system only gets new tasks after both finishes.

So if you have one task of 2 hours and another of 12 hours, you will have to wait 12 hours to get new tasks. It seems they have not tasks enough to all crunchers that are making FreeHAL.

Fire$torm
08-01-12, 04:29 PM
Sorry Z, I haven't run any recently. I have some now and will report what I find.

Fire$torm
08-01-12, 11:16 PM
OK Z, I'm seeing what you are. If the credits are the same, as in fixed per wu and not based on run time, then FH just got really lame....

Mumps
08-01-12, 11:37 PM
Well, looking at the daily output at FreeHal for the last month, it's fairly easy to see where those WU times probably increased. Because the daily output there has dropped to less than half of what it was 2 weeks ago.

Duke of Buckingham
08-02-12, 06:55 AM
I have a computer with low memory that when was running FreeHAL put the other tasks waiting for memory. I used the task manager and FreeHAL was taking almost 1Gb of memory for running. I had to stop FreeHAL at that computer.

artemis8
08-05-12, 01:01 AM
Yeah, mine are taking much much longer as well.

denim
08-08-12, 10:43 PM
On my quad-core at work, I am steady getting 3 WU's finish in normal time, and the 4th running really long. :(

spingadus
10-11-12, 03:54 PM
Seeing this on my box as well. 6 of 8 finish early. The last 2 take another 4 hours. So, i'm losing 4 hours of work for the 6 other tasks, since no new tasks will download until all 8 have completed. Aborting the long tasks might start a new download of tasks, but that takes a lot of micro managing. Of course, then you would lose the work done. Probably not worth the effort. I can't see any way of choosing tasks size in the prefs.

Mike029
10-11-12, 09:27 PM
Seeing this on my box as well. 6 of 8 finish early. The last 2 take another 4 hours. So, i'm losing 4 hours of work for the 6 other tasks, since no new tasks will download until all 8 have completed. Aborting the long tasks might start a new download of tasks, but that takes a lot of micro managing. Of course, then you would lose the work done. Probably not worth the effort. I can't see any way of choosing tasks size in the prefs.

Just another reason I stopped running this. I also found that FH was slowing down my gpus.

Bryan
10-12-12, 12:24 AM
I'm surprised that the guy that was tauted as a "boy genius" can't figure out how to issue the same sized WU on a single work request. Brilliant, absolutely brilliant. I too have stopped running the project.

John P. Myers
10-12-12, 12:46 AM
Got one going now that's expected to finish in 2 hours for a total runtime of 10 hours. Also noticed it's taking about 256MB RAM to run it, where a single Poem GPU app uses half that. This will be my last FH.

zombie67
10-12-12, 08:31 AM
Hmm. I don't see any slow-down running FH. To me, it's all just free upside..

Bryan
10-12-12, 11:27 AM
Hmm. I don't see any slow-down running FH. To me, it's all just free upside..

There are 2 different sized WU being issued by the project. If you get 7 of the shorter ones and 1 or more of the longer ones then all the cores that had crunched the short WU sit idle for a couple of hours waiting for the last WU to finish. It won't ask for work until ALL WU have completed.

Yes, somewhat free credits, but no where near what you could get if the WU size was matched on each download.

zombie67
10-12-12, 12:34 PM
There are 2 different sized WU being issued by the project. If you get 7 of the shorter ones and 1 or more of the longer ones then all the cores that had crunched the short WU sit idle for a couple of hours waiting for the last WU to finish. It won't ask for work until ALL WU have completed.

Yes, somewhat free credits, but no where near what you could get if the WU size was matched on each download.

I understand. But free is free. It's not like I am losing out on credits from other projects, by running FH. Even with mixed length tasks. It's all upside. For me, it's ~2500 per day, and over 14M in total, helping out the team's total score.

I'm surprised that this isn't a priority for the folks who want us to be #1 in overall credits.

John P. Myers
10-12-12, 01:16 PM
Hmm. I don't see any slow-down running FH. To me, it's all just free upside..

Since ending FH i've noticed Poem WUs complete about 7%-10% faster. Fluke? Maybe. I'll keep a watch on it and see if that trend continues.

However Mike029 reports the same thing, though not sure of his increase %

Mike029
10-12-12, 01:42 PM
Since ending FH i've noticed Poem WUs complete about 7%-10% faster. Fluke? Maybe. I'll keep a watch on it and see if that trend continues.

However Mike029 reports the same thing, though not sure of his increase %

When I stopped running FH it knocked off 3 minutes off of my Poem wu's FRom 44 mins to 41 mins. Perhaps it due to me running an T1045 vs an I7.

Duke of Buckingham
10-12-12, 02:42 PM
I don't know what is the problem but the FreeHAL tasks are getting bigger and bigger, my other computer stopped running them after a small while and was "waiting for memory" all the time and this one when running NFS 16e start losing time maybe because of the big amount of swap needed. I think it depends on the Computer memory, Operational System and the projects that we are running.

zombie67
10-12-12, 04:11 PM
Yes, FH tasks require a lot of RAM. And I think it increases over the duration of the task. So longer tasks need more RAM by the end. This is why I have been recommending for *years* that people should have at *least* 2gb per thread, for any cruncher. There are always a few projects that need lots of RAM, especially the medical projects. So don't skimp. I target 4gb per thread. This comes in very handy when running VMs. So I can run FH tasks in the VM, as well as natively.