I try to find on the threads and can´t find the answer.
I want to crunch the AP WU ASAP after they DL, anyone knows how to set a configuration do that? I don´t want to wait to crunch the entire cache of MB for start the crunching of the AP and leave the MB on cache for to be crunched when no AP WU where avaiable.

---

I try to find on the threads and can´t find the answer.
I want to crunch the AP WU ASAP after they DL, anyone knows how to set a configuration do that? I don´t want to wait to crunch the entire cache of MB for start the crunching of the AP and leave the MB on cache for to be crunched when no AP WU where avaiable.

Ya, I'd like to know that too. From what I see with all the knobs and switches in BOINC, there doesn't appear to be a way to do that - at least not from the GUI.

This makes me think back to my days of administering large IBM mainframe systems. If only BOINC had the level of detailed control that mainframe job scheduling had.

---

You could suspend all of the MB tasks in your queue one at a time, which will allow the AP tasks to process. The only problem with this is that the scheduler will not d/l any more work because the task manager has tasks suspended. You then have take the time to resume the MB tasks, one at a time, and hope that AP tasks will be d/l'ed. Like you I'd rather do AP then MB, but I guess until there is a steady stream of AP being split and sent, you might as well do what I'm doing -- suck it up and deal with it.

---

I don't buy computers, I build them!!

I know we could do that manualy, but requires a lot of babysitting, what i want is an automatic way to to that. Something easy to do with serveral hosts.

---

I know we could do that manualy, but requires a lot of babysitting, what i want is an automatic way to to that. Something easy to do with serveral hosts.

I don't think the powers that be will allow us mere peasants to do that. You could set up a separate incidence of BOINC and only run AP on it.

---

I don't buy computers, I build them!!

Running another instance of Boinc running AP will not do what i want to do. Keep allways running 1 AP + 1MB (the optimal performance on my hosts) and when no AP avaiable run 2 MB. There must be a way to tell the boinc to change the order of what is the next wu to crunch.

---

I think it would be possible to write a DOS script to basically be a scheduler on top of a scheduler. It could suspend and resume tasks as the 5 minute time approached. This would not be trivial because of timing issues, but it would allow for the possibility of doing what you're looking for.

I've given thought to writing a scheduler on top of the Boinc Gui that would be tailorable to whatever type of requirements you can imagine. This might include varying the number of concurrent tasks on an individual GPU basis.

Perhaps it would be better to take the source code for the Boinc manager and make changes to it at that level. All you would rely upon Berkeley is to supply you with work units every 5 minutes. All things on your client machine would be only limited by your imagination and your ability to implement it in code. Unfortunately, I just don't have the time for such an undertaking now.

---

I belive that´s could be a way to do, i was thinking to build a small program that could monitoring the cache and make the needed changes, just don´t know the structure of this file or where to find the info.

On other hand i don´t want to loose time developing something if somebodycould have a easy solution, that why i ask for help if anyone knows a way easely to do that.

---

Running another instance of Boinc running AP will not do what i want to do. Keep allways running 1 AP + 1MB (the optimal performance on my hosts) and when no AP avaiable run 2 MB. There must be a way to tell the boinc to change the order of what is the next wu to crunch.

Actually there may be, but no automatic way, unless you write some code. You could edit the file after it is on your hard drive and change the due date/time to something shorter for the AP's. That should make BOINC go into EDF mode for that W/U.

I think the separate instance of BOINC, with a separate data directory may be the right way to attack this. You also need to create a second account at Seti. On each instance tell it to only use half of the CPU's. On one account tell it to only do MB on the other tell it to only do AP. On each you can enable do work for another application if nothing is available. As long as they don't find out about each other you should be able to have each happily using their 1/2 of your CPU.

This won't quite do exactly what you want, but it as close as your are likely to get.

---

If i want to do the code, did anyone knows the name of the file boinc uses to choose what is the next WU to crunch and where i can find it´s structure?
I belive that´s is the best aproach to the problem.

Another thing that will be interesting if i build the code is a way to make the cache mantain a certain number of MB and AP to keep the host working with 1 AP+1MB all the possible time.

---

what i want is an automatic way to to that. Something easy to do with serveral hosts.

Possibly 2 instances of Boinc, set to run APs and the other to run MBs on their respective processors.

---

I try to find on the threads and can´t find the answer.
I want to crunch the AP WU ASAP after they DL, anyone knows how to set a configuration do that? I don´t want to wait to crunch the entire cache of MB for start the crunching of the AP and leave the MB on cache for to be crunched when no AP WU where avaiable.

On your Seti@Home account page, under Preferences for this project, you can set one or more crunchers to preferentially run AP by setting your preferences to:
S@H Enhanced: no
S@H V7: no
Astropulse v6: yes
If no work for selected applications, allow work from other applications? yes

That sets you up to run AP when they are available,, and MB v7 when AP is not available.

If you only want to do this on 1 machine, set it as a separate localtion (default, home, work, school) from the others.

---

Donald
Infernal Optimist / Submariner, retired

2 instances will not work, why? it still have the cache sequence of what is the next wu to crunch to meet. What i need is a way to bypass this order.

---

Donald

I allready to that, but that´s make the host crunch the entire old MB cache before start to crunch the new AP when avaiable, that´s not what i want, my ideia is start the AP crunche ASAP after the WU is dl.

---

Donald

I allready to that, but that´s make the host crunch the entire old MB cache before start to crunch the new AP when avaiable, that´s not what i want, my idea is start the AP crunche ASAP after the WU is dl.

That being the case, you might have to do the manual suspension that was mentioned earlier. Once the AP starts, you can unsuspend the MBs. You might also want to adjust your cache settings to only have 1 or 2 APs onboard at a time, to minimize the number of MBs downloaded during an AP drought.

---

Donald
Infernal Optimist / Submariner, retired

Yes i manualy do that actualy now too, but that requires a lot of babysitting and is a pain in the a... if you have to make the same thing on several hosts...

I think the only solution is build the program to change the sequence automaticaly, will do some more digin in that direction lets see what i can do.

Thanks all for the help.

---

May I ask why you want to run that way?

The way the project is currently set up, there may be a possibility to twist BOINC's arm. I'll PM.

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

May I ask why you want to run that way?

The way the project is currently set up, there may be a possibility to twist BOINC's arm. I'll PM.

Simply... at least on my multiple GPU hosts if i crunch 1 AP and 1 MB on each one of the GPU is far more eficient (by eficient i means, crunch more WU and produces more credit in less time). For example if you crunch 6 AP at the same time on a 3xGPU I5 host because you have no free core to feed the GPU´s all slow down, so the best performance is to allow 3 AP (that uses 3 cores) and 3 MB (that uses the other core)

Will take a deep look at your PM.

<edit> The other problem is the way the cache work on Boinc, some kind of FIFO, so for example if i have the cache filled with MB and it receive a AP, to start to crunch that WU you need to wait the entire MB cache to be crunched (several hours even on a multiple fast GPU host). As the way the splitters are working on this days, not all the time you have AP to crunch but there are plenity of MB, so we need to find a way to crunch AP as fast as they where dl in order to receive more wu before the server stops to have this AP avaible to DL. By doing that your RAC will be optimized...

OK i know the credit "means nothing stuff", but is like a orchestra, the rouler dictates the compass, and since our rouler (who controls the creditnew stuff) is deaf to our shoutings (asking for balance the credit paid AP & MB to a similar gains), we the musicians need to try to find a way to extract the best from our instruments.

---

The only way to do what you want is two instances of BOINC, each attached to Seti using a different account. Then you can add the RAC's by hand.

If you have to have this under one account, then you can download the BOINC tarballs and have a go at modifying the scheduler. At present the scheduler doesn't even know about MB's and AP's.

From the project's perspective MB's are just as important as AP's.

---

Running 2 instances of Boinc will have the same problem i will need to wait the old cache to be crunched to start to crunch the recent DL AP WU.

From the project's perspective MB's are just as important as AP's

Don´t want to enter in any "agressive negotiations" but...
If that is realy the case, then why give 2x more credit to a AP than MB for the same processing time/power used?

Yes i could be wrong, but i worry about credit, and i know i´m not the only one. Or for what other reason i want to worry about what WU to crunch? For years i crunch only MB (my choice i know, but i have a good reason, MB does not need too many free cores as AP and that is good for my I5 multi GPU´s hosts who has very few cores avaiables) and never worry about that credit diference (MB vs AP) but after V7 that diference is..... forget about, that discusion is for another thread.

---