1 task of each app???
log in

Advanced search

Message boards : Number crunching : 1 task of each app???

Author Message
Profile Nosferatu*
Send message
Joined: 26 Nov 15
Posts: 8
Credit: 17,230,850
RAC: 3,725
Message 612 - Posted: 29 Mar 2016, 23:35:05 UTC
Last modified: 29 Mar 2016, 23:35:38 UTC

My pc's only receive one task of each type work unit at this time. One of V1 and one of V2. I have noticed that there are some computers with 50 or more tasks running at the same time. That's like 25 or more of each app. How is this possible? It seems that this project has a limit set of one task per app per host pc at one time. Am I wrong in my assumptions or is there a way to get and run more than one task at a time?

Profile Nosferatu*
Send message
Joined: 26 Nov 15
Posts: 8
Credit: 17,230,850
RAC: 3,725
Message 613 - Posted: 29 Mar 2016, 23:37:25 UTC

Update. Just found pc running 73 tasks at one time.

Profile Steve Hawker*
Send message
Joined: 10 Nov 15
Posts: 70
Credit: 1,491,750
RAC: 2,911
Message 614 - Posted: 30 Mar 2016, 4:48:23 UTC - in response to Message 612.

My pc's only receive one task of each type work unit at this time. One of V1 and one of V2. I have noticed that there are some computers with 50 or more tasks running at the same time. That's like 25 or more of each app. How is this possible? It seems that this project has a limit set of one task per app per host pc at one time. Am I wrong in my assumptions or is there a way to get and run more than one task at a time?
\r\n\r\nI've never tried it but multiple VMs might do it although I thought each VM would appear as a single device.\r\n\r\nAdmin has taken action on VMs before and I guess we'll see if action will be taken again.

Profile Nosferatu*
Send message
Joined: 26 Nov 15
Posts: 8
Credit: 17,230,850
RAC: 3,725
Message 615 - Posted: 30 Mar 2016, 7:05:48 UTC

This is one machine is running many tasks at the same time. Don't know if it is legit or not. Seems there are at least several machines running in this manner.\r\n\r\nIf I figure things correctly then one task V1 and V2 run in just over an hour.\r\nThe 2 get you 20 points. In 24 hrs that would be 450 to 460 points. So how are so many getting well above 1000 points a day recent average credit? The best you could get per host would be around 460 or 470 average credit.\r\n\r\nNeed to find out limit on task at once per host. I am only able to download one of each at a time, and no additional tasks in buffer seem to be able to be downloaded.

Profile Eric_Kaiser
Send message
Joined: 3 Feb 16
Posts: 12
Credit: 4,164,480
RAC: 10,146
Message 618 - Posted: 30 Mar 2016, 18:23:29 UTC - in response to Message 615.
Last modified: 30 Mar 2016, 18:25:14 UTC

So how are so many getting well above 1000 points a day recent average credit?
\r\nThe answer is simple. I have 10 physical distinct computers running linux and one windows computer. Credit sums up to ~5200 points per day.\r\n\r\nAnd I know at least one member who has a cluster of odroid arm devices running linux. If I remember correctly he has more than 40 of them...

Profile krzyszp
Send message
Joined: 9 Nov 15
Posts: 14
Credit: 1,473,869
RAC: 2,752
Message 619 - Posted: 30 Mar 2016, 18:40:08 UTC - in response to Message 618.
Last modified: 30 Mar 2016, 18:40:36 UTC

So how are so many getting well above 1000 points a day recent average credit?
\r\nThe answer is simple. I have 10 physical distinct computers running linux and one windows computer. Credit sums up to ~5200 points per day.\r\n\r\nAnd I know at least one member who has a cluster of odroid arm devices running linux. If I remember correctly he has more than 40 of them...
\r\nI think, this is the cluster:\r\nhttp://forum.odroid.com/viewtopic.php?f=93&t=18120\r\n\r\n;)
____________

HAL9000
Avatar
Send message
Joined: 19 Nov 15
Posts: 88
Credit: 66,085,300
RAC: 146,392
Message 621 - Posted: 31 Mar 2016, 1:28:33 UTC - in response to Message 618.

So how are so many getting well above 1000 points a day recent average credit?
\r\nThe answer is simple. I have 10 physical distinct computers running linux and one windows computer. Credit sums up to ~5200 points per day.\r\n\r\nAnd I know at least one member who has a cluster of odroid arm devices running linux. If I remember correctly he has more than 40 of them...
\r\nI think they were asking how so many hosts, not users, have a RAC of over 1000.\r\n\r\nThe answer is those hosts are running more than 1 task per app per client. I would assume those users modified their BOINC client to ignore the limit of 1 task per app for NCI tasks.
____________

Join the BP6/VP6 User Group today!

Profile Nosferatu*
Send message
Joined: 26 Nov 15
Posts: 8
Credit: 17,230,850
RAC: 3,725
Message 622 - Posted: 31 Mar 2016, 4:25:23 UTC - in response to Message 621.

HAL9000,\r\n\r\nYou are correct. That is what I am asking. I also thought that the limit was set by the project itself not boinc. An old retired boinc project name FreeHAL also ran NCI. We were able to run one wu per host. I seem to remember that they later set it to give one task per core. Don't know if my memory is correct or not but I seem to remember that they started out with one task per host and changed it to one task per core. My memory might be mixed up on the order but I do recall running it both ways. Think they even had project settings that allowed running on per host or one per core.\r\nGuess I am still trying to figure out how to do it and if project allows this to be done without repercussions.\r\n\r\nThanks HAL9000.\r\n\r\nCould a project admin please advise.

Profile Moises Cardona
Send message
Joined: 6 Dec 15
Posts: 91
Credit: 7,227,900
RAC: 11,429
Message 623 - Posted: 31 Mar 2016, 11:52:30 UTC - in response to Message 622.

HAL9000,\r\n\r\nYou are correct. That is what I am asking. I also thought that the limit was set by the project itself not boinc. An old retired boinc project name FreeHAL also ran NCI. We were able to run one wu per host. I seem to remember that they later set it to give one task per core. Don't know if my memory is correct or not but I seem to remember that they started out with one task per host and changed it to one task per core. My memory might be mixed up on the order but I do recall running it both ways. Think they even had project settings that allowed running on per host or one per core.\r\nGuess I am still trying to figure out how to do it and if project allows this to be done without repercussions.\r\n\r\nThanks HAL9000.\r\n\r\nCould a project admin please advise.
\r\n\r\nI think FreeHal at a moment added regular CPU intensive task. I remember If I'm not mistaken that there was a project with both CPU and non-CPU Intensive tasks but I don't remember if that was freehal at all

Profile Eric_Kaiser
Send message
Joined: 3 Feb 16
Posts: 12
Credit: 4,164,480
RAC: 10,146
Message 625 - Posted: 31 Mar 2016, 12:31:59 UTC

It is possible to run multiple boinc instances on a computer. Each instance needs its own data directory.\r\nThis is all I know about it. Never diven into it any deeper.

Profile Eric_Kaiser
Send message
Joined: 3 Feb 16
Posts: 12
Credit: 4,164,480
RAC: 10,146
Message 626 - Posted: 31 Mar 2016, 12:35:44 UTC - in response to Message 619.
Last modified: 31 Mar 2016, 12:36:57 UTC

\r\nI think, this is the cluster:\r\nhttp://forum.odroid.com/viewtopic.php?f=93&t=18120\r\n
\r\nYes, that is only one of his odroid clusters. He also has one cluster with all odroids watercooled. If I remember correctly the watercooled cluster was in one of the hardkernel magazines.\r\nAnd he is building up one with the odroids c2:\r\nhttp://abload.de/image.php?img=dsc0006269a9z.jpg

HAL9000
Avatar
Send message
Joined: 19 Nov 15
Posts: 88
Credit: 66,085,300
RAC: 146,392
Message 629 - Posted: 1 Apr 2016, 3:13:44 UTC - in response to Message 622.

HAL9000,\r\n\r\nYou are correct. That is what I am asking. I also thought that the limit was set by the project itself not boinc. An old retired boinc project name FreeHAL also ran NCI. We were able to run one wu per host. I seem to remember that they later set it to give one task per core. Don't know if my memory is correct or not but I seem to remember that they started out with one task per host and changed it to one task per core. My memory might be mixed up on the order but I do recall running it both ways. Think they even had project settings that allowed running on per host or one per core.\r\nGuess I am still trying to figure out how to do it and if project allows this to be done without repercussions.\r\n\r\nThanks HAL9000.\r\n\r\nCould a project admin please advise.
\r\nThe server has set that the apps are NCI. Which tells the BOINC client to treat them differently from standard CPU/GPU intensive tasks. When the client sees the NCI flag is set it says "OK I only ever need 1 task form this app/project. If the BOINC client were modified to ignore the NCI flag then it would treat the tasks like normal try to fill the queue.\r\n\r\nThere is an option in the BOINC server to limit the number of tasks. A strict limit can be implemented on a per host, per CPU, per CPU core, or per GPU basis. I'm not sure if there is an option to limit per application.\r\nIf the admin were planning on only ever having two apps at once then a hard limit of 2 tasks per host could be set. That would slow down those that are getting more than 1 task per app. Until they use the work around for the server limits. Which also exists & doesn't even require modifying the client.
____________

Join the BP6/VP6 User Group today!

Profile Conan
Send message
Joined: 9 Nov 15
Posts: 39
Credit: 1,218,133
RAC: 2,478
Message 648 - Posted: 3 Apr 2016, 12:22:36 UTC

I have seen before on some other projects that a single computer can present itself to a BOINC project as the "gateway" or "master" for a cluster that is sitting behind that computer. \r\nI am not sure how that gets accomplished but others with far more networking knowledge should have little trouble setting that up.\r\nOnly the one computer contacts the project and then downloads and distributes work for the cluster members behind it.\r\n\r\nThat could be what is happening here.\r\nFor instance, a host showing 40 tasks may have 20 'nodes' behind it (or 19 nodes if the master computer is also processing data).\r\n\r\nConan

Profile Jaska*
Send message
Joined: 7 Apr 16
Posts: 2
Credit: 512,750
RAC: 450
Message 700 - Posted: 7 Apr 2016, 17:07:12 UTC

The admin of FreeHAL did not take kindly to people using cc_config (or app_data, which came later) to show <ncpus> as a number greater than logical threads. He saw it as "cheating" and even went so far as to manually remove people's credit. Needless to say, the project shut down, in no small part because of his asinine idea of "project management."\r\n\r\nNewer versions of BOINC now permit app_info XMLs in order to run tasks as anonymous platform, but in most cases (especially with optimised apps) this is not necessary.\r\n\r\nI am not sure about modifying these files to run one or more task per core. I would try, but doubt it would do much good - plus, the project admin may have something to say about how such practice could affect app activity. \r\n\r\nFor example, running more than one instance of the QCN nci app simultaneously would be pointless. \r\n\r\nIt's interesting, though, that my hosts with 4, 8, 8, and 12 threads are each processing only two tasks at a time. Perhaps it's that this is still a relatively new project.

Profile Moises Cardona
Send message
Joined: 6 Dec 15
Posts: 91
Credit: 7,227,900
RAC: 11,429
Message 702 - Posted: 7 Apr 2016, 17:29:22 UTC - in response to Message 700.

The admin of FreeHAL did not take kindly to people using cc_config (or app_data, which came later) to show <ncpus> as a number greater than logical threads. He saw it as "cheating" and even went so far as to manually remove people's credit. Needless to say, the project shut down, in no small part because of his asinine idea of "project management."\r\n\r\nNewer versions of BOINC now permit app_info XMLs in order to run tasks as anonymous platform, but in most cases (especially with optimised apps) this is not necessary.\r\n\r\nI am not sure about modifying these files to run one or more task per core. I would try, but doubt it would do much good - plus, the project admin may have something to say about how such practice could affect app activity. \r\n\r\nFor example, running more than one instance of the QCN nci app simultaneously would be pointless. \r\n\r\nIt's interesting, though, that my hosts with 4, 8, 8, and 12 threads are each processing only two tasks at a time. Perhaps it's that this is still a relatively new project.
\r\n\r\nThat's because you have 1 task per app. Since this project has 2 apps, you receive 1 task per app per host running the project

Profile [boinc@poland]goofyx
Project administrator
Project developer
Project scientist
Help desk expert
Send message
Joined: 8 Nov 15
Posts: 542
Credit: 9,002,054
RAC: 8,283
Message 705 - Posted: 7 Apr 2016, 18:48:09 UTC - in response to Message 702.

Monkeys_v1 and Monkeys_v2 is non-cpu project it is mean that they do not care about number of your cpu core/thread because they always processing 1WU per application.

Profile Jaska*
Send message
Joined: 7 Apr 16
Posts: 2
Credit: 512,750
RAC: 450
Message 707 - Posted: 7 Apr 2016, 20:34:46 UTC

Right. I thought OP was under the impression people with very high RAC and/or hosts processing larger numbers of WUs were using app_info XMLs to tell BOINC to run more WUs for this project than just 2 at a time. \r\n\r\nApologies if I misunderstood. Perhaps there has been some update to BOINC's handling of NCI applications of which I was unaware.

Profile Moises Cardona
Send message
Joined: 6 Dec 15
Posts: 91
Credit: 7,227,900
RAC: 11,429
Message 708 - Posted: 7 Apr 2016, 22:37:20 UTC - in response to Message 707.

Right. I thought OP was under the impression people with very high RAC and/or hosts processing larger numbers of WUs were using app_info XMLs to tell BOINC to run more WUs for this project than just 2 at a time. \r\n\r\nApologies if I misunderstood. Perhaps there has been some update to BOINC's handling of NCI applications of which I was unaware.
\r\n\r\nThere are some people who have clusters made of Raspberry Pi's/Orange Pi's/ODroid and other devices which is why some have very high RAC, due to each cluster processing the 2 apps at once per machine in the cluster :)

HAL9000
Avatar
Send message
Joined: 19 Nov 15
Posts: 88
Credit: 66,085,300
RAC: 146,392
Message 709 - Posted: 8 Apr 2016, 20:58:56 UTC - in response to Message 708.

Right. I thought OP was under the impression people with very high RAC and/or hosts processing larger numbers of WUs were using app_info XMLs to tell BOINC to run more WUs for this project than just 2 at a time. \r\n\r\nApologies if I misunderstood. Perhaps there has been some update to BOINC's handling of NCI applications of which I was unaware.
\r\n\r\nThere are some people who have clusters made of Raspberry Pi's/Orange Pi's/ODroid and other devices which is why some have very high RAC, due to each cluster processing the 2 apps at once per machine in the cluster :)
\r\nUnless those users have overcome the shared memory issue that BOINC uses to communicate with the science apps I would expect they are running an instance of BOINC on each node in their cluster. Even if someone were to overcome the issue & have a single instance of BOINC across their cluster. They would still be limited to 1 NCI application per instance of BOINC.\r\n\r\nThe OP's question is about those users that are causing BOINC to run more than 1 NCI application per BOINC instance. \r\nGiven tasks run ~1hr, are 10 credits, & there are currently 2 apps. It should then not be possible for a single host to have a RAC much higher than about 480. However if you look at the Top Hosts list. You will see 100's of hosts well over than. With their task list showing 3 or more tasks in progress for each app. BOINC is designed to not allow that. However since anyone can download the source and compile their own client it isn't that hard to get around limits.\r\n\r\nEven if the project were to specifically set a limit on the number of tasks in progress that can be gotten around as well. It is likely the same trick those users are using on their hosts to have more tasks.
____________

Join the BP6/VP6 User Group today!

Message boards : Number crunching : 1 task of each app???


Main page · Your account · Message boards


Copyright © 2020 Goofyx

Generated 4 Jun 2020, 8:36:36 UTC