BOINC has went a long way since SAH started.

Having 3 projects (2 steady: SAH and CPDN and one whenever it's needed: LHC) I would like to prioritise tasks in BOINC.

By prioritising I mean 2 things:

1. by task: I would like to set the "crunching" order for my tasks in a project. Id est: I receive 10 or 20 tasks, and looking through them I want to have the AP WU that came in last, running first, since I know with my resource share (below this list), other tasks in the project will be taken care of; but I don't want to suspend the other tasks just to get to that one task.
   
   
2. by project: this one is a little bit more complicated to explain though (or mabe not?). Setting processor affinity for project tasks. Id est: I have a dual core processor running a task in each core. looking at the TaskManager, usually CPDN runs at pretty much 50% (99% on one core) while SAH is given only 30-45% on the other core sharing it with most other programs. What I would like is to set the priority for a project, so that in my case SAH uses the 50% and CPDN is given the 30-40%. (LHC doesn't matter since whenever they pop in, and I notice them, I give them full priority)
   

Note: this is not the resource share for a project, for which I have set:
CPDN (80) 16%
SAH (100) 20%
LHC (320) 64%

Are my 2 priority wishes even possible? I know the first one could be, but it would require some programming in BOINC.

---

Robi

BOINC has went a long way since SAH started.

Having 3 projects (2 steady: SAH and CPDN and one whenever it's needed: LHC) I would like to prioritise tasks in BOINC.

By prioritising I mean 2 things:

1. by task: I would like to set the "crunching" order for my tasks in a project. Id est: I receive 10 or 20 tasks, and looking through them I want to have the AP WU that came in last, running first, since I know with my resource share (below this list), other tasks in the project will be taken care of; but I don't want to suspend the other tasks just to get to that one task.
   
   
2. by project: this one is a little bit more complicated to explain though (or mabe not?). Setting processor affinity for project tasks. Id est: I have a dual core processor running a task in each core. looking at the TaskManager, usually CPDN runs at pretty much 50% (99% on one core) while SAH is given only 30-45% on the other core sharing it with most other programs. What I would like is to set the priority for a project, so that in my case SAH uses the 50% and CPDN is given the 30-40%. (LHC doesn't matter since whenever they pop in, and I notice them, I give them full priority)
   

Note: this is not the resource share for a project, for which I have set:
CPDN (80) 16%
SAH (100) 20%
LHC (320) 64%

Are my 2 priority wishes even possible? I know the first one could be, but it would require some programming in BOINC.

Rumour has it that some major changes will be made to Boinc's scheduler functions in the coming weeks........months....whatever.
Mostly to cope with the complexity of Cuda's impact on the scheduler and how it calculates things....

I don't think that you will ever be able to select what tasks you wish to run first........it's too inherent in the way Boinc functions....and even if you could, at some point Boinc would have to take over to keep everything in balance...

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

Rumour has it that some major changes will be made to Boinc's scheduler functions in the coming weeks........months....whatever.
Mostly to cope with the complexity of Cuda's impact on the scheduler and how it calculates things....

I don't think that you will ever be able to select what tasks you wish to run first........it's too inherent in the way Boinc functions....and even if you could, at some point Boinc would have to take over to keep everything in balance...

I've noticed that BOINC takes over anyway when a task reaches its deadline anyway :)
up until now I've been suspending certain tasks I want done at a later time, but it would be much easier if I could just click and drag a task to the top and BOINC rewrites the client_state.xml fie accordingly.
besides, I have also noticed that after running BOINC for a while, and I restart the computer, the order of projects/tasks is rewritten.
the only big thing to implement in BOINC would be the click'n'drag or drag&drop and if BOINC keeps the priority for deadline tasks, then the balance would still be maintained.

in a way, I'm glad I don't have CUDA on this rig, although I've been toying with the idea... but reading about the problems, and the ability to fry the GPU I've been rethinking the installation.

I just hope someone at the development level reads this thread and maybe, just maybe, we'll have a "prioritised BOINC" :) ;) wink wink, nudge nudge

---

Robi

I don't think that you will ever be able to select what tasks you wish to run first........it's too inherent in the way Boinc functions....and even if you could, at some point Boinc would have to take over to keep everything in balance...

It also brings in an inherent danger.

BOINC has an internal simulation where it "runs" all of the work in the queue, in order, honoring resource shares, etc. and checks deadlines to make sure everything is met. If something is at risk of missing the deadline, BOINC will run that work in high-priority and get it done (and in) on time.

So, let's take Robi's example. He has lots of MB, and a single AP unit. He wants to push the MB ahead of the AP work. A bunch of MB gets finished, and we get down to the point of having just a couple of MB units. BOINC tops-up the queue, and Robi again manually reshuffles the order.

BOINC can find itself in a position where it has a 60 hour AP unit with a deadline just two days away.

I understand that people really want to reshuffle the queue. If one can resist, one will find that BOINC does a good job doing work in order assigned, with enough exceptions to make sure deadlines aren't missed.

Remember too that the BOINC scheduler is supposed to deal with work that takes anywhere from a few minutes, to more than a year. Whenever I think I've noticed something that'd help the scheduler, JM7 reminds me about CPDN.

---

I don't think that you will ever be able to select what tasks you wish to run first........it's too inherent in the way Boinc functions....and even if you could, at some point Boinc would have to take over to keep everything in balance...

It also brings in an inherent danger.

BOINC has an internal simulation where it "runs" all of the work in the queue, in order, honoring resource shares, etc. and checks deadlines to make sure everything is met. If something is at risk of missing the deadline, BOINC will run that work in high-priority and get it done (and in) on time.

So, let's take Robi's example. He has lots of MB, and a single AP unit. He wants to push the MB ahead of the AP work. A bunch of MB gets finished, and we get down to the point of having just a couple of MB units. BOINC tops-up the queue, and Robi again manually reshuffles the order.

BOINC can find itself in a position where it has a 60 hour AP unit with a deadline just two days away.

I understand that people really want to reshuffle the queue. If one can resist, one will find that BOINC does a good job doing work in order assigned, with enough exceptions to make sure deadlines aren't missed.

Remember too that the BOINC scheduler is supposed to deal with work that takes anywhere from a few minutes, to more than a year. Whenever I think I've noticed something that'd help the scheduler, JM7 reminds me about CPDN.

I've seen Boinc switch to a new WU with a short deadline, I mean It takes what? 1.5hrs or so for My 3.51GHz Quad PC to crunch one and the Deadline could be one to two weeks out, Still It shifts to what Boinc thinks is a higher priority WU deadline wise.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

the only big thing to implement in BOINC would be the click'n'drag or drag&drop and if BOINC keeps the priority for deadline tasks, then the balance would still be maintained.

BOINC does not keep the "priority" for deadline tasks, and the client state file is not in any particular order.

As a general rule, BOINC processes work in the order it was assigned by the project(s).

It periodically runs a simulation based on the projected CPU time, the system uptime, duration correction factors, and deadlines.

If the simulation shows a work unit "in danger" then that work unit runs "high priority" until it is out of danger.

Any imbalance caused by running "high priority" is tracked through "long term debt" and corrected later (when new work is downloaded).

Is it perfect? No, but it does work pretty well.

Want to have some fun? Stop BOINC and fiddle with duration_correction_factor in the client_state.xml file. Make it big and you'll see BOINC run everything in "high priority" mode. Make it small and BOINC will over-request work and (eventually) work out of the mess surprisingly well.

If you do decide to mess with some of these values, keep in mind that if you break it, you own both parts.

---

I don't think that you will ever be able to select what tasks you wish to run first........it's too inherent in the way Boinc functions....and even if you could, at some point Boinc would have to take over to keep everything in balance...

It also brings in an inherent danger.

BOINC has an internal simulation where it "runs" all of the work in the queue, in order, honoring resource shares, etc. and checks deadlines to make sure everything is met. If something is at risk of missing the deadline, BOINC will run that work in high-priority and get it done (and in) on time.

So, let's take Robi's example. He has lots of MB, and a single AP unit. He wants to push the MB ahead of the AP work. A bunch of MB gets finished, and we get down to the point of having just a couple of MB units. BOINC tops-up the queue, and Robi again manually reshuffles the order.

BOINC can find itself in a position where it has a 60 hour AP unit with a deadline just two days away.

I understand that people really want to reshuffle the queue. If one can resist, one will find that BOINC does a good job doing work in order assigned, with enough exceptions to make sure deadlines aren't missed.

Remember too that the BOINC scheduler is supposed to deal with work that takes anywhere from a few minutes, to more than a year. Whenever I think I've noticed something that'd help the scheduler, JM7 reminds me about CPDN.

I've seen Boinc switch to a new WU with a short deadline, I mean It takes what? 1.5hrs or so for My 3.51GHz Quad PC to crunch one and the Deadline could be one to two weeks out, Still It shifts to what Boinc thinks is a higher priority WU deadline wise.

The code behind that is fairly paranoid, and I think for good reason.

What BOINC doesn't know about are outages (power or connectivity), and it doesn't know if the estimates (the current DCF) is accurate.

BOINC wants to finish work with plenty of time to upload and report.

So, it starts crunching "priority" to finish with time to spare. There are fudge-factors and allowances to make up for the fact that we can't always predict the future.

---

The reason why the order is changed when you stop and restart is that you have run the tasks for some time. When BOINC Manager starts it looks at the tasks using the simulation Ned mentioned and then selects the tasks to be run ... which is almost never the ones that *HAD* been running, because the internal numbers have changed.

One of the rules that frustrates me to no end is the task that has 30 seconds to 5 minutes to run to completion and BOINC suddenly starts running something else ... I always wanted a rule that would check to see if the task was within my guidelines and not switch out ...

Yeah, I know, Orbit, where the end few minutes took days ... so that is the exception, next pass, if the progress and time to complete has not changed THEN you can suspend ...

Anyway, with 4 to 8 cores and the shares you selected I would say that you would get a fair shake at doing almost what you want. The only fly in the ointment is that LHC almost never has work and so ... hit or miss to get tasks ... and when they do have work it is gone in moments ... and I find that very annoying as that is by far the project I want to work on ... atom smashing ... what could be more fun?

---

I think in general you won't be seeing more micro-management functions for individual WUs integrated into BOINC beyond the suspend/abort buttons already in existence. That's because doing more micro-managing than necessary can very often lead to BOINC running WU's that it wouldn't otherwise run. You can prevent the debts from stabilizing and as a result BOINC won't be able to run with the resource shares you set. You can also force it into high priority situations more often than usual, again unstabilizing the debts and resource shares.

---

I think in general you won't be seeing more micro-management functions for individual WUs integrated into BOINC beyond the suspend/abort buttons already in existence. That's because doing more micro-managing than necessary can very often lead to BOINC running WU's that it wouldn't otherwise run. You can prevent the debts from stabilizing and as a result BOINC won't be able to run with the resource shares you set. You can also force it into high priority situations more often than usual, again unstabilizing the debts and resource shares.

I partially agree, but by suspending tasks you can accomplish the same, even take it as far as just, lets say, intentionally suspend tasks like AP and then just as the deadline arrives let the task run while it is given to someone else.
Since the task is already running, a "cancel" will not be accepted and the extra task returned, thus using unnecessary CPU cycles :)

My Idea, by micro managing them locally, I could place an AP task on the top (that's what I would like to do since I have 2 cores), keep AP crunching while CPDN and the other SAH tasks share the rest of the CPU allotted time.

I do understand that there are weak systems out there that just barely manage to keep tasks "on time".

While experimenting with my system, I have been able to "prioritise" tasks with the Task Manager, setting the CPDN task to CPU0 and SAH task to CPU1, the latter using the full 99-100% of the second core, but neat would be to let BOINC allow me to prioritise the tasks I want running tops.
My CPDN task claims to be done in 4 months, of course, giving it a little less then half the CPU time between it and SAH, it will take a little over 8 months, where 4 months have already passed and 60% of the WU has been completed (and I still have over two years for the deadline #8->.

The point is that I know how my system runs and how I could manage the tasks for them to smoothly run side by side and still finish on time.

Of course, as everybody has been pointing out, BOINC manages it by itself with a few user tweaks...
Still, it would be nice :)

---

Robi

Doesn't this all incorrectly ASSUME that returning results 'early' makes a difference? As long as your computer is processing WUs and retruning them prior to the deadline, what does it matter?

I'm a big fan of "Set it & Forget it"

---

SETI@home since 1999
"Set it, and Forget it!"

While experimenting with my system, I have been able to "prioritise" tasks with the Task Manager, setting the CPDN task to CPU0 and SAH task to CPU1, the latter using the full 99-100% of the second core, but neat would be to let BOINC allow me to prioritise the tasks I want running tops.
My CPDN task claims to be done in 4 months, of course, giving it a little less then half the CPU time between it and SAH, it will take a little over 8 months, where 4 months have already passed and 60% of the WU has been completed (and I still have over two years for the deadline #8->.

Setting the processor affinity, in independent testing, has shown no overall performance gain. In fact, locking the prcesses to a particular processor can actually lower performance, or at least lower the responsiveness of the system while using it, which is a major taboo with BOINC in trying to be so user friendly and unintrusive.

The point is that I know how my system runs and how I could manage the tasks for them to smoothly run side by side and still finish on time.

You may know how your system runs, but micromanaging defeats the purpose of the built-in scheduler, and it defeats the purpose of designing it to require little or no human interaction.

Of course, there are some here who prefer to micromanage because they want BOINC to behave a certain way, but I don't think its very time efficient to code two different scheduling designs into BOINC (one for micromanagers and one for the set-it-and-forget-it types) when the majority of users are the latter. ...and I'm still of the opinion that if the micromanagers simply ignored BOINC, with the exception of checking it every once in a while, would behave much less like Obsessive Compulsive Disorder (not meant in an offensive way), and wouldn't feel the need to control every aspect of the software when its already designed to do it for you.

[ ...and I'm still of the opinion that if the micromanagers simply ignored BOINC, with the exception of checking it every once in a while, would behave much less like Obsessive Compulsive Disorder (not meant in an offensive way), ...

Oh, Yeah?

Well, my OCD took offense ...


:)

Sorry, I could not resist ....

---

[ ...and I'm still of the opinion that if the micromanagers simply ignored BOINC, with the exception of checking it every once in a while, would behave much less like Obsessive Compulsive Disorder (not meant in an offensive way), ...

Oh, Yeah?

Well, my OCD took offense ...


:)

Sorry, I could not resist ....

---

BOINC Wiki . . .

Science Status Page . . .

[ ...and I'm still of the opinion that if the micromanagers simply ignored BOINC, with the exception of checking it every once in a while, would behave much less like Obsessive Compulsive Disorder (not meant in an offensive way), ...

Oh, Yeah?

Well, my OCD took offense ...


:)

Sorry, I could not resist ....

LOL

[ ...and I'm still of the opinion that if the micromanagers simply ignored BOINC, with the exception of checking it every once in a while, would behave much less like Obsessive Compulsive Disorder (not meant in an offensive way), ...

Oh, Yeah?

Well, my OCD took offense ...


:)

Sorry, I could not resist ....

LOL

Well at least My Anxiety, Concentration, Thyroid, Joint problems, Back Problems, Weight, Eyesight, Arthritis and Depression don't take offense, It must be the Depression that's in charge right now.

I was informed of a county program that could pay the closing costs and the deposit of up to $50,000.00, Now If only I can talk to David @ the USDA(fixed 33yr mortgage) and get more details on the county program from My Real Estate agent, the idea of have a mortgage with insurance that costs less than $100 is simply too good to pass up.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Well, I read through the entire thread, and I am hoping someone will point out that I missed something. Because it looks like to me that no one pointed out the real problem with the way BOINC operates in regards to switching WU's to high priority.

It shouldn't do it at all.

In fact it is rather terrible that it changes priority without warning. In fact, I think it is sort-of a violation of the trust people place in the BOINC programmers and the whole concept of sharing your computer for grid computing purposes.

I just ran into the problem this thread discusses. Suddenly, all of my eight processors were co-opted by BOINC. Why? Well, I didn't know. After all, it hadn't happened before. So I had to spend time, something I have had to do a lot with regarding BOINC, unfortunately.

Of course, I checked all the options. Limit this, limit that, limit everything EXCEPT when BOINC just up and decides that it needs to steal your CPU power from every other task in the system because it thinks (in my case hilariously in error) that it might not finish a WU in time (OMYGOD! WHOLLY KRAP. STOP THE WORLD I NEED TO HOP OFF.)

I think the BOINC people don't really consider their audience. Maybe they don't care about anyone who does this just out of interest in helping. Some real-life figures: Running SETI costs me about $14.00 a month in extra electricity usage. This is easy to measure with simple devices. Literally, running all 8 processors for SETI causes the power consumption of my Mac Pro 8-core machine to increase by that much. Also, running processors at full bore constantly increases their runtime temperature, dramatically (40 degrees Farenheit or more) which almost certainly reduces their lifetime.

And most of all, BOINC sucks up my time as well as my CPU cycles. I want BOINC to work, but it seems to be a challenge-a-week. And EVERY REQUEST to "experts" who are supposedly out there to help have gone unanswered.

I know that the BOINC apologists out there will be ready and willing to tell me that if I "just adjust parameter X" and "in this situation, adjust parameter Y", and "what's wrong with you, it's all explained quite clearly at this web page", and "if you just take the projected completion time of this WU and divide by the total number of WU's assigned, and then take the square root and multiply it by E=Mc2, then you would just be fine and dandy!"... and would you just get a grip on everything and etc. etc. etc. DONE.

BOINC steals my CPU time, not to mention my personal time. I can't let it. The cost to me of my time alone should have made me give up on this a long time ago.

I signed up to Seti@Home (and later BOINC when I was forced to) not to incur one headache after another. The usability of the BOINC manager program is NIL. As in, crap. As in, what the hell were they thinking? If it requires such special knowledge to get it to work, it's a wonder anyone participates.

I may give it another week or so for some answers to come, either from this post, or the (I think now non-existent) experts out there. And after that, I am going to save $14.00 a month, CPU stress, and human stress.

---

Set your amount of CPUs to use to 50% (that's 4 on your 8 CPU system, leaving you with 4 free for other things) either in your web site preferences or if you use the Local Advanced Preferences instead, set it there.

It's the preference "On multiprocessor systems, use at most xxx% of the processors"

---

Yes, that is basically the make-shift work-around kind of thing I've already done.

Unfortunately, it is no answer at all to the problem I having with BOINC. It is a coerced work-around that ignores the fact that I have two other non-BOINC processes that utilize all eight processors.

---

In that case your choices are to:

- Not run BOINC at all. Abort all work, uninstall the program and be on your merry way.
- Set BOINC to only do work when the CPUs are idle.
- Set BOINC to run only during certain times, or certain times on certain days, when you're reasonably sure those other processes aren't taking up the CPUs.

Oh and they aren't workarounds, but features of the BOINC system, to make it easier for you to decide how to use the program.

A workaround would be to write a batch file specifically for your computer that stops BOINC completely when the other process begins and restarts BOINC after it's done.

---