Message boards :
Number crunching :
I Thought BOINC Was Crazy But...
Message board moderation
Author | Message |
---|---|
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
Sheesh! On one of my machines BOINC is running a bunch of APs due 4/21 - not in high priority mode. Meanwhile, it has stopped ("Waiting to Run") 3 APs due 4/10. Also waiting are another bunch in "ready to Start" that are due 4/11 - 4/17. Similarly, my other machine has 2 APs Waiting-to-Run due 4/9, and a bunch running due 4/21 also. Say what? Can anyone explain the logic of this to me under LIFO, FIFO or FEFIFOFUM? |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Imagine I'm your boss (scary thought, I know). Imagine I gave you some major work to do that I want on my desk by 4/10, which is two weeks away. You begin the project in earnest and get so far before I show up again and give you another major project that I want on my desk by 4/21, which is three and a half weeks away. Somewhere in there I also gave you some work that I want on my desk by 4/11 through 4/17 respectively. With the amount of work I've given you, you begin to wonder if you'll get it all done on time. So instead of just focusing on the ones with the earliest deadlines while possibly not having enough time afterward to complete the other tasks I gave you, you decide to dedicate small amounts of time to each task. Knowing that if you balance your time well, you'll be able to complete all tasks on time without having to stress yourself out, and change your priorities to Earliest Deadline First mode, which shouldn't be necessary given the amount of time due inbetween tasks. That's all BOINC is doing. It's slowly giving CPU time to each task to ensure that it doesn't need to run in earliest deadline first / high priority mode. |
Helli_retiered Send message Joined: 15 Dec 99 Posts: 707 Credit: 108,785,585 RAC: 0 |
That's a very fine explanation, $o |>0/|/3. :-) Helli |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
Interesting. I will check the elapsed times on the "Waiting to Run" 4/10 APs and see if they are really being run in a round-robin fashion with the 4/21s. |
perryjay Send message Joined: 20 Aug 02 Posts: 3377 Credit: 20,676,751 RAC: 0 |
Wait, you mean BOINC isn't crazy after all? But that would mean I'm.... Oh dear!!! :-( PROUD MEMBER OF Team Starfire World BOINC |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
Well, I went out for about 4 hours. When I came back, the "Waiting to Run" APs were exactly where they were before, to the second. So I don't think they are being run alternately with the later ones that are "Running" now. So that theory is shot down. NOW can someone explain the behavior of running ones due later while stopping ones due earlier? Isn't that counter-intuitive, not to mention WRONG? After all, if BOINC is afraid that the 4/21 APs aren't going to finish in time, why does he seem to think that the 4/10 ones will? PS: Perry, take a deep breath and sit down. Don't get excited. Think about NICE things. |
Mike Send message Joined: 17 Feb 01 Posts: 34256 Credit: 79,922,639 RAC: 80 |
I´m wondering about this behaviour for years. Specially for APs on ATI it makes no sense. The guys think way to complicated. It will finnish exactly the same amount of work if running in FIFO. You only have to wait longer for your wingman. Running on huge caches its better to suspend WUs with later deadlines. So i get what i want. Maybe in 20 years those guys will wake up. With each crime and every kindness we birth our future. |
JohnDK Send message Joined: 28 May 00 Posts: 1222 Credit: 451,243,443 RAC: 1,127 |
The few times I've run Seti + backup projects with short deadlines, some of the backup work wouldn't have finished in time if I didn't pause the Seti WUs which Boinc insisted on running. Have no idea why it happens. |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
Well, I went out for about 4 hours. When I came back, the "Waiting to Run" APs were exactly where they were before, to the second. So I don't think they are being run alternately with the later ones that are "Running" now. I'm afraid your time sample may be too short. You may want to observe over a much longer time period. NOW can someone explain the behavior of running ones due later while stopping ones due earlier? Isn't that counter-intuitive, not to mention WRONG? After all, if BOINC is afraid that the 4/21 APs aren't going to finish in time, why does he seem to think that the 4/10 ones will? BOINC runs many simulations based upon available computing time and amount of time the computer is running and the cached workload on hand and resource share. If it has sufficiently processed enough work from the 4/10 ones but now the simulation has shown that the 4/21s are in danger, it will start processing those, balancing the CPU time between all the aforementioned variables. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
If you want to know why BOINC makes the decisions it does then you should enable the logging flags that display the scheduler output. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
If you want to know why BOINC makes the decisions it does then you should enable the logging flags that display the scheduler output. OK - how do I do that? |
Mike Send message Joined: 17 Feb 01 Posts: 34256 Credit: 79,922,639 RAC: 80 |
I'm afraid your time sample may be too short. You may want to observe over a much longer time period. It didn´t learn after 500 units i´ve finnished so i dont expect it would after 1 million. It could be so easy. With each crime and every kindness we birth our future. |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
I'm afraid your time sample may be too short. You may want to observe over a much longer time period. Interesting... it works as designed for me. It is amazingly easy. |
Mike Send message Joined: 17 Feb 01 Posts: 34256 Credit: 79,922,639 RAC: 80 |
I'm afraid your time sample may be too short. You may want to observe over a much longer time period. It does so long you dont have a huge cache. Running a 20 day cache is only possible in FIFO mode. I finnished 512 APs in 20 days with GPU. Yes it is possible to get such a cache. With each crime and every kindness we birth our future. |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
If you want to know why BOINC makes the decisions it does then you should enable the logging flags that display the scheduler output. Do you know about the cc_config.xml file? SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
Cruncher-American Send message Joined: 25 Mar 02 Posts: 1513 Credit: 370,893,186 RAC: 340 |
If you want to know why BOINC makes the decisions it does then you should enable the logging flags that display the scheduler output. Tes I did, but not about debugging flags. I will investigate... Thanks! |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
I'm afraid your time sample may be too short. You may want to observe over a much longer time period. Then don't run a large cache. I would fully support a decision to modify the code server-side that restricts the cache to 5 days max for all projects. |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
Then don't run a large cache. I would fully support a decision to modify the code server-side that restricts the cache to 5 days max for all projects. A 10 day max I would certainly support but not 5. Cheers. |
OzzFan Send message Joined: 9 Apr 02 Posts: 15691 Credit: 84,761,841 RAC: 28 |
I received a PM making the following suggestion to my original analogy: You're my boss (shudder!) and you say "have these tasks on my desk by (deadline), each one will take 10 hours." My original analogy didn't properly convey DCF as this suggestion does. |
Mike Send message Joined: 17 Feb 01 Posts: 34256 Credit: 79,922,639 RAC: 80 |
I'm afraid your time sample may be too short. You may want to observe over a much longer time period. Wouldn´t change anything. You can run a 40 day cache if you know how. I dont need server code for that. Anyhow it seems you missed the point. Running in FIFO mode would only make it easier. But i know some guys like it complicated. One of the reasons i quit Boinc alpha. With each crime and every kindness we birth our future. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.