log in

Multiple tasks enabled


Advanced search

Message boards : News : Multiple tasks enabled

1 · 2 · 3 · 4 . . . 5 · Next
Author Message
Profile Ben Segal
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 1 Nov 10
Posts: 1162
Credit: 264
RAC: 0
Message 17319 - Posted: 7 Oct 2014, 12:04:52 UTC

After many volunteer requests over the years, we have decided it is time to send out multiple T4T tasks to users who enable more than 1 CPU in their BOINC preferences. To begin with we have enabled 2 tasks maximum.

Volunteer feedback about this change is welcome - please use this News thread.

The Team

Cphipps
Send message
Joined: 1 Aug 14
Posts: 10
Credit: 617,850
RAC: 1,010
Message 17325 - Posted: 7 Oct 2014, 15:11:41 UTC - in response to Message 17319.

Thanks

Darth Beaver
Avatar
Send message
Joined: 14 Jun 13
Posts: 17
Credit: 9,583
RAC: 0
Message 17326 - Posted: 7 Oct 2014, 16:06:12 UTC

good idea i have 8 cores on 1 machine and the other has 4 and even my Lappy has 2 so about time , and the GPU can do at least 3 units on the other projects so how about using the GPU's too .

And i will add why do you use a VM shorely you can write a standalone app and i'm shore it would be faster than using a VM and be more compatible with slower machines and less complicated to use for newb's
____________

Profile Tom95134
Send message
Joined: 23 Mar 11
Posts: 356
Credit: 348,330
RAC: 0
Message 17327 - Posted: 7 Oct 2014, 16:16:01 UTC
Last modified: 7 Oct 2014, 16:20:32 UTC

When my system "woke up" this morning (I shutdown overnight) it was happily running two vLHC tasks. So far I have no problems.

However, it appears that the number of vLHC tasks running in controlled by the settings of the Server and not (as far as I can see) by any BOINC or Project Preferences. Once I saw that two vLHC Tasks were running I went to the Preferences and changed the Processors from "--" (unlimited) to 2. Read the Preferences and all Tasks were suspended except for the 2 vLHC Tasks.

So, since I like to be able to "service" the original LHC Project whenever they have Tasks to run (and two other projects) can someone provide a more detailed explanation on how to use the Resource Share (and other settings that are not very clear how they are to be used) so that some percentage of the CPUs (I have 8 cores) will run other Tasks should you ever choose to enable more than twp Tasks?

Thanks.


Windows 7 Professional x64, INTEL i7-2600 3.40GHz, 16GB RAM, NVDIA GeForce GTS-450

Crystal Pellet
Volunteer moderator
Avatar
Send message
Joined: 5 Aug 11
Posts: 1561
Credit: 1,425,548
RAC: 22
Message 17329 - Posted: 7 Oct 2014, 17:05:03 UTC

Number of connected machines raised from 811 at 01:30 CEST this morning to 1394 at 18:30 CEST :-)

Profile Ben Segal
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 1 Nov 10
Posts: 1162
Credit: 264
RAC: 0
Message 17330 - Posted: 7 Oct 2014, 17:07:06 UTC - in response to Message 17327.

Tom, you shouldn't change the number of processors downwards like that. If it's only 2 then only two tasks will run and so if your resource shares are set to favour T4T, all your other projects' tasks will stall. If you want two T4T's to go most of the time, set your number of CPU's to 2+X where X is the average number of other tasks you'd like to run also.

In the Number Crunching thread I see you are asking about resource shares, so over to that…

Ben

Profile tullio
Send message
Joined: 28 Nov 10
Posts: 1540
Credit: 1,456,969
RAC: 1,716
Message 17331 - Posted: 7 Oct 2014, 17:11:29 UTC

I received two vLHC@home tasks on my Windows PC with 8 GB RAM running 2 ATLAS@home tasks and one VLHC task,and immediately one of the ATLAS tasks was waiting for memory. So I suspended one of the vLHC tasks. It seems that 2 ATLAS tasks and one vLHC is the limit with 8 GB. I need more RAM.
TULLIO

Profile Ben Segal
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 1 Nov 10
Posts: 1162
Credit: 264
RAC: 0
Message 17332 - Posted: 7 Oct 2014, 17:12:39 UTC - in response to Message 17329.

Number of connected machines raised from 811 at 01:30 CEST this morning to 1394 at 18:30 CEST :-)

Yep, we are watching this. I think it's OK. Depending on how many people run with multiple CPUs we could get up to 2x the number of running VM's that we had before. The number should hopefully level off soon.

(We are checking our CoPilot servers right now in case some are down, as this can cause a rise in the number of connected VMs).

Thanks a lot, CP, always solid support from you!

Ben

Profile Chilean
Avatar
Send message
Joined: 29 Nov 12
Posts: 84
Credit: 331,351
RAC: 0
Message 17333 - Posted: 7 Oct 2014, 17:15:43 UTC
Last modified: 7 Oct 2014, 17:31:25 UTC

Sweet. I should be able to reach the billion mark faster now.

EDIT: Why not let us choose? And leave the default @ 1 or 2?
____________

SuperSluether
Avatar
Send message
Joined: 3 Sep 14
Posts: 16
Credit: 6,594
RAC: 0
Message 17337 - Posted: 7 Oct 2014, 18:45:33 UTC - in response to Message 17333.

Finally, no more messing around with multiple clients! Now the only problem is the same old co-exist problem with other VMs, but I think that sorted itself out on my computer.

BTW, a few people were asking about the VMs. I can't say for certain, but I think vLHC and ATLAS get most of their working files from a network on CERN, which also explains why they won't run without internet access.
____________

Profile Tom95134
Send message
Joined: 23 Mar 11
Posts: 356
Credit: 348,330
RAC: 0
Message 17338 - Posted: 7 Oct 2014, 18:51:26 UTC

I'm not sure about this but it looks like once vLHC captures a CPU it doesn't release it (swap out) as other BOINC Tasks do. I vaguely remember something about this but is this the normal operation of BOINC/VM?

SuperSluether
Avatar
Send message
Joined: 3 Sep 14
Posts: 16
Credit: 6,594
RAC: 0
Message 17339 - Posted: 7 Oct 2014, 19:06:02 UTC - in response to Message 17338.

I'm not sure about this but it looks like once vLHC captures a CPU it doesn't release it (swap out) as other BOINC Tasks do. I vaguely remember something about this but is this the normal operation of BOINC/VM?


I've seen vLHC swap out before. Sometimes projects switch out every hour, other times the same project will run all day. I think it's related to how BOINC manages resource shares between projects, and not specific to the projects themselves.
____________

Profile Ray Murray
Volunteer moderator
Avatar
Send message
Joined: 10 Aug 11
Posts: 224
Credit: 1,807,415
RAC: 2,723
Message 17340 - Posted: 7 Oct 2014, 19:14:24 UTC - in response to Message 17338.

Hi Tom,
Boinc will take maybe a few hours to a day to relearn the new settings with 2 tasks running.
Boinc work scheduling and resource share is an (overly) complicated beast which sometimes defies logic, especially when something gets changed. It will often give 1 project more time than seems necessary until another projects drops below its share threshold before swapping over. Over a long period the resource shares will average out to whatever values you have set but over a shorter time scale it may not appear to work as one would expect.

SuperSluether
Avatar
Send message
Joined: 3 Sep 14
Posts: 16
Credit: 6,594
RAC: 0
Message 17341 - Posted: 7 Oct 2014, 19:30:31 UTC - in response to Message 17340.

Hi Tom,
Boinc will take maybe a few hours to a day to relearn the new settings with 2 tasks running.
Boinc work scheduling and resource share is an (overly) complicated beast which sometimes defies logic, especially when something gets changed. It will often give 1 project more time than seems necessary until another projects drops below its share threshold before swapping over. Over a long period the resource shares will average out to whatever values you have set but over a shorter time scale it may not appear to work as one would expect.


This may not be the place to ask, but how does BOINC schedule resource shares? Is it determined by credit? Work (FLOPS) done? Time per project?
____________

Crystal Pellet
Volunteer moderator
Avatar
Send message
Joined: 5 Aug 11
Posts: 1561
Credit: 1,425,548
RAC: 22
Message 17342 - Posted: 7 Oct 2014, 19:40:46 UTC - in response to Message 17340.

Boinc work scheduling and resource share is an (overly) complicated beast which sometimes defies logic, especially when something gets changed. It will often give 1 project more time than seems necessary until another projects drops below its share threshold before swapping over. Over a long period the resource shares will average out to whatever values you have set but over a shorter time scale it may not appear to work as one would expect.

Very off topic!
But (for Windows users), if you are familiar with DOS commands open a command box (cmd from the start menu),
change your directory to BOINC DATA and give the commands:

copy client_state.xml temp.txt
find "</master_url>" temp.txt
find "</rec>" temp.txt

You get a list of your projects and a list in the same order of the actual recent average credit of the projects above calculated by BOINC.
The project with the lowest REC will run first, except when a task could not make the deadline.

Profile Mumak
Avatar
Send message
Joined: 14 Feb 14
Posts: 44
Credit: 3,720,191
RAC: 3,644
Message 17343 - Posted: 7 Oct 2014, 20:14:43 UTC

A very nice addition :)
How does this work with the <avg_ncpus>x</avg_ncpus> in app_config ?
Say I get now 2 VMs running and when setting avg_ncpus to 2 will that result in utilizing 4 threads ?

captainjack
Send message
Joined: 13 Sep 11
Posts: 60
Credit: 586,296
RAC: 0
Message 17344 - Posted: 7 Oct 2014, 20:22:23 UTC

I started up two tasks on my Linux box this morning. After 4 hours 16 minutes tasks number 2 says "Waiting to run (Scheduler wait: VM Job unmanageable, restarting later)". However, the System Monitor shows that both VBoxHeadless processes are using a full CPU each.

Running Virtualbox 4.3.12 and Ubuntu 14.04.

Let me know if you need more information.

Profile MAGIC Quantum Mechanic
Avatar
Send message
Joined: 1 Mar 11
Posts: 662
Credit: 8,491,510
RAC: 7,722
Message 17345 - Posted: 7 Oct 2014, 21:02:15 UTC

To me I think 2 tasks is just right (like it was the first time) since we also have Atlas and LHC tasks just for Cern and most members run other projects too.

I basically only run Einstein GPU tasks besides these Cern tasks.

I have 5 hosts back to running vLHC X2 again ......guess I will get to work on bringing host #7 back to life and seeing if I can get host #6 to get back to *recognizing* its GPU card so it can do those for Einstein and get back to vLHC X2 again.

(can't wait to see what the credits turn out to be running X2 here)

That will also decide which way I will do this here.


-Samson
____________

Profile Ben Segal
Project administrator
Project developer
Project tester
Project scientist
Send message
Joined: 1 Nov 10
Posts: 1162
Credit: 264
RAC: 0
Message 17346 - Posted: 7 Oct 2014, 21:06:34 UTC - in response to Message 17344.

I started up two tasks on my Linux box this morning. After 4 hours 16 minutes tasks number 2 says "Waiting to run (Scheduler wait: VM Job unmanageable, restarting later)". However, the System Monitor shows that both VBoxHeadless processes are using a full CPU each.

Running Virtualbox 4.3.12 and Ubuntu 14.04.

Let me know if you need more information.

I would abort task number 2: that is the dreaded vboxwrapper error when it loses control of the VM. You may even have to kill the vboxheadless process of that task if aborting doesn't do it for you (make sure you don't kill the wrong one…).

Ben

Profile Toby Broom
Volunteer moderator
Send message
Joined: 28 Feb 11
Posts: 350
Credit: 9,942,429
RAC: 15,298
Message 17347 - Posted: 7 Oct 2014, 23:16:47 UTC

This could take some working out for me, I have 1 pc running 8 instances of BOINC to do 8 WU's, I need to drop this to 4.

:-~

1 · 2 · 3 · 4 . . . 5 · Next

Message boards : News : Multiple tasks enabled