Advanced search

Message boards : Number Crunching : Running multiple tasks on a multi-core system takes longer.

Author Message
Profile Charles Dennett
Avatar
Send message
Joined: 10 Aug 14
Posts: 39
Combined Credit: 354,259
DNA@Home: 219,783
SubsetSum@Home: 134,472
Wildlife@Home: 4
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4890 - Posted: 15 Dec 2014, 20:11:54 UTC

One of my crunchers is a 4 core 64 bit Linux box, Typically I was giving equal time to SIMAP and DNA@Home which meant I was running 2 SIMAP tasks and 2 DNA@Home tasks at the same time. I've run out ot SIMAP work as that project is shutting down and is now out of work. So, this cruncher is now running 4 DNA@Home tasks at a time.

I've noticed a funny thing. DNA@Home tasks now take about twice as long with 4 tasks at a time as when only 2 are running. In other words, a task that would take 40 seconds when DNA@Home used 2 cores now takes 80 seconds when all 4 cores are used for DNA@Home. Likewise a 6 minute task now takes 12 minutes. Why would that be?

Now the question is, with the doubled time, will the credit double or stay the same?

-Charlie
____________

Profile Charles Dennett
Avatar
Send message
Joined: 10 Aug 14
Posts: 39
Combined Credit: 354,259
DNA@Home: 219,783
SubsetSum@Home: 134,472
Wildlife@Home: 4
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4891 - Posted: 15 Dec 2014, 21:33:45 UTC

Running some tests. I've limited DNA@Home to only 1 core. No other project is crunching. So far it's only completed a couple of small snail (10fa) tasks. With two cores they used to take about 44 seconds. The two done with only 1 core in use took 30 seconds. It's working on a large snail (1000fa) now. Normally that take 1 hour 7 minutes. It competed 10% in 4 minutes 42 seconds which if my math is correct extrapolates to 47 minutes for the whole thing.

-Charlie
____________

Profile Charles Dennett
Avatar
Send message
Joined: 10 Aug 14
Posts: 39
Combined Credit: 354,259
DNA@Home: 219,783
SubsetSum@Home: 134,472
Wildlife@Home: 4
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4899 - Posted: 16 Dec 2014, 13:15:23 UTC

Using the snail_hg19_1000fa_1motif type of task, I've run these with 1,2,3 and 4 cores on my system. Only these tasks were running. No other boinc projects were active. Total run time per task is just a bit longer than cpu time per task.

1 core: Takes about 46 minutes each . Get about 21 credits
2 cores: Takes about 1 hour 7 minutes each - Gets about 21 credits
3 cores: Takes about 1 hour 40 minutes each - Gets about 22 credits
4 cores: Takes about 2 hours 29 minutes each - Gets about 23 credits

The system is running Fedora 20 with a KDE desktop but it's not actively used most of the day so close to 100% of the cpu is used for DNA@Home.

Is this normal behavoir? I have not run the same tests on other projects. Perhaps I will. Now that SIMAP is done I might go back to POEM and see how that behaves (but I will come back to DNA@Home after those tests.)

-Charlie
____________

Travis Desell
Volunteer moderator
Project administrator
Project developer
Project scientist
Send message
Joined: 16 Jan 12
Posts: 1813
Combined Credit: 23,514,257
DNA@Home: 293,563
SubsetSum@Home: 349,212
Wildlife@Home: 22,871,482
Wildlife@Home Watched: 212,926s
Wildlife@Home Events: 51
Climate Tweets: 21
Images Observed: 774

              
Message 4901 - Posted: 16 Dec 2014, 14:31:31 UTC - in response to Message 4899.

Using the snail_hg19_1000fa_1motif type of task, I've run these with 1,2,3 and 4 cores on my system. Only these tasks were running. No other boinc projects were active. Total run time per task is just a bit longer than cpu time per task.

1 core: Takes about 46 minutes each . Get about 21 credits
2 cores: Takes about 1 hour 7 minutes each - Gets about 21 credits
3 cores: Takes about 1 hour 40 minutes each - Gets about 22 credits
4 cores: Takes about 2 hours 29 minutes each - Gets about 23 credits

The system is running Fedora 20 with a KDE desktop but it's not actively used most of the day so close to 100% of the cpu is used for DNA@Home.

Is this normal behavoir? I have not run the same tests on other projects. Perhaps I will. Now that SIMAP is done I might go back to POEM and see how that behaves (but I will come back to DNA@Home after those tests.)

-Charlie


I don't know if this is normal, but the binaries might be compiled to use some hardware which might be why they are slower if multiple WUs are running at the same time? They might be using some vector multiplication or something along those lines. That's the only thing I can think of. Very strange...

yo2013
Send message
Joined: 27 Sep 14
Posts: 19
Combined Credit: 963,744
DNA@Home: 118,965
SubsetSum@Home: 122,043
Wildlife@Home: 722,736
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

      
Message 4903 - Posted: 16 Dec 2014, 15:23:36 UTC - in response to Message 4901.

Maybe there is a HD bottleneck. Try running iotop to see disk usage.

Profile Charles Dennett
Avatar
Send message
Joined: 10 Aug 14
Posts: 39
Combined Credit: 354,259
DNA@Home: 219,783
SubsetSum@Home: 134,472
Wildlife@Home: 4
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4904 - Posted: 16 Dec 2014, 16:13:44 UTC

Disk activity on the system is very low. It only has 4 GB of memory, the most the MB will support. Could be memory pressure. Also, just one piece of silicon with 4 cores so it could be some shared component on the processor chip or other hardware. Anyway, running tests now with another BOINC app (POEM) to see how it behaves. Just started tests with a one task at a time. It will be a while. I'll report back.

-Charlie
____________

Amauri
Send message
Joined: 12 Aug 14
Posts: 2
Combined Credit: 364,558
DNA@Home: 2,603
SubsetSum@Home: 0
Wildlife@Home: 361,955
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

  
Message 4905 - Posted: 16 Dec 2014, 20:32:57 UTC - in response to Message 4904.

These tasks use a lot of memory (600 MB), so I'm sure it's a memory bottleneck.

Profile Charles Dennett
Avatar
Send message
Joined: 10 Aug 14
Posts: 39
Combined Credit: 354,259
DNA@Home: 219,783
SubsetSum@Home: 134,472
Wildlife@Home: 4
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4906 - Posted: 16 Dec 2014, 21:44:02 UTC

Looks like the same problem does not exist for the POEM project. Tasks take the same amount of time no matter how many cores I use.

When I get some time I can play with a kernel setting called vm.swappiness. It adjusts when the kernel starts using swap space. If my memory is correct, the default is 60. (Ranges 0-100). 60 means that it will start using swap space when 60% of the memory is free. Setting it lower will allow it to use more memory before it starts using swap.

-Charlie
____________

gtippitt
Send message
Joined: 26 Mar 14
Posts: 4
Combined Credit: 10,688,594
DNA@Home: 950,688
SubsetSum@Home: 0
Wildlife@Home: 9,737,906
Wildlife@Home Watched: 0s
Wildlife@Home Events: 0
Climate Tweets: 0
Images Observed: 0

    
Message 4955 - Posted: 14 Jan 2015, 16:48:39 UTC

Your problem is most likely that you don't have enough RAM for more than 1 or 2 CSG/DNA tasks to run concurrently. This project requires a great deal more RAM per tasks than POEM. The discussion in the "Memory Usage" thread in this forum has details on this and how to balance the load on multi-core systems.

Some of the DNA tasks currently running on my systems right now require 1.5 GB per tasks, which for your systems would be about half the systems' memory for 1 DNA task. Restricting your systems to running 1 or 2 DNA tasks, and allowing POEM to use the other cores will give you better overall throughput of work units.

Good Luck,
Greg


Post to thread

Message boards : Number Crunching : Running multiple tasks on a multi-core system takes longer.