Prioritise projects rather than share

gravywavy
gravywavy
Joined: 22 Jan 05
Posts: 392
Credit: 68962
RAC: 0
Topic 189401

BOINC's approach to multiple projects at present is one of sharing resource rather than prioritising. If you want to make sure the machine keep crunching through a SETI outage, you run 20% of your resource for Einstein, etc.

This is not quite what I want.

What I *really* want is to give 100% of my time to my favoured project; if that is unavailable to then give my time to the next favoured project, and so on.

How I'd see this working is that normally the work cache would be full of work for the favoured project, and only that work run. Work would be asked for and this would replenish the cache. Other projects would not get a look in so long as the priority project was supplying work.

When work was asked for and refused, or if the server was unreachable, then an equivalent amount of work would be asked for from the second priority project. If that failed then the third priority, etc.

Once downloaded, the BOINC client would ensure that it was run in time to be returned for its deadline. Within that boundary projects which had work on the local machine would be allocated resources according to work share as at present. With the new deadline watching facilities in the laterst versions of BOINC, I think this would be workable.

It would mean no change to the scheduling of switching between projects, the project priority would instead act as a 'veto' when asking for more work.

There would be a new parameter in the prefs, project priority, 1 for the main project, 2 for the next, and so on. The default would be for all projects to be priority 1, so that if the prefs were left unaltered things would go on exactly as at present.

The two questions are, whether it better meets participants' wishes than the current scheme, and secondly if so whether it is feasible from a programming point of view.

I can only speak for the first: I would much prefer to give all my machine time to my favoured project, and only run the backup when my favorite had not supplied any work. What do other participants feel?

~~gravywavy

[AF>Quebec] louchio
[AF>Quebec] louchio
Joined: 23 Feb 05
Posts: 6
Credit: 54118
RAC: 0

Prioritise projects rather than share

I totaly agree whit your statement. Je suis totalement d'accord avec votre suggestion

Blank Reg
Blank Reg
Joined: 18 Jan 05
Posts: 228
Credit: 40599
RAC: 0

It works that way now, all I

It works that way now, all I do is tell Boinc not to let XYZ@H download any WUs and if my Fav goes down, I tell my backup(s) to get busy and get some work. I use CC 4.45 and I like the No New WORK button.
I only have 2 of my 4 boxes running just one project and they are Alpha, so this helps me keep track of any problems. The other 2 crunch multiple projects as it should be.....

Sharky T
Sharky T
Joined: 19 Feb 05
Posts: 159
Credit: 1187722
RAC: 0

RE: It works that way now,

Message 13275 in response to message 13274

Quote:
It works that way now, all I do is tell Boinc not to let XYZ@H download any WUs and if my Fav goes down, I tell my backup(s) to get busy and get some work. I use CC 4.45 and I like the No New WORK button.

That needs a human eye to check the state all the time.
I think he wants it automatic..

BTW,I think I seen a thread where the primary project resource was set to 10000
and the backup project set to 1.(can't find it now)
Now, if I remember correctly,what happends is that the first time it downloads WU for each project it probably goes into EDFstate(panic mode) for the backup project and finishes it,uploads it,and will NOT download a new WU for it until primary project is out of work and cant get any new ones.
(not as pretty as buttons for prioritising ;)


Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5872
Credit: 117234512098
RAC: 36091169

I think this suggestion has a

I think this suggestion has a lot of merit and caters for the wishes of the majority of crunchers. Most would have a "favourite" project and most would see it as a fun thing to put their other backup projects in a preferred order. Then, if they got sick of that priority order, they could rearrange it from time to time. It also caters for people who want to try out a new project for a week or two and then revert if they wished. For example, if you wanted to do a burst on the new Seti Test project, you could if you made it your #1 priority this week and then simply reset your priorities if you wished to go back to the previous order. JM7's new scheduler code might have to sort out the mess but it's smart enough to do that :).

However the Devs are probably not going to like it because it allows the user to concentrate on just a single project rather than spreading the load. There have been statements from time to time which indicate that the BOINC plan is to encourage the users to adopt multiple projects rather than prioritizing on mainly one.

However, from a selfish point of view, I would very much support this type of prioritizing system and trust that individual users will have different favourites and thus the load will be spread anyway.

Well done on an extremely worthwhile suggestion!! (Gravywavy for President!!!! :).)

Cheers,
Gary.

gravywavy
gravywavy
Joined: 22 Jan 05
Posts: 392
Credit: 68962
RAC: 0

RE: However the Devs are

Message 13277 in response to message 13276

Quote:

However the Devs are probably not going to like it because it allows the user to concentrate on just a single project rather than spreading the load. There have been statements from time to time which indicate that the BOINC plan is to encourage the users to adopt multiple projects rather than prioritizing on mainly one.

yes, and the reason always quoted is that this allows machine cycles to be saved when a project runs out of work. My scheme does the same.

It also lessens the load on databases, as with fewer projects with local wu, over the whole community there might well be fewer wu booked out at any one time. Hopefully this angle might appeal to the devs, especially the E@H devs who seem particularly concerned about database pressure.

...and thanks for the warm feedback

~~gravywavy

Bronco
Bronco
Joined: 22 Jun 05
Posts: 6
Credit: 73429
RAC: 0

I'm not sure this way of

I'm not sure this way of making PC work is logic as long as we have short dead lines. This would mean having a lot of WU going over, and would increase server's work.

I think that we are suffitiently able to choose, in the current way, what we want to do. Having coefs of 950 and 50 means 95% of the workload for one project, which is quite the result wanted, and have the second project in backup in case.

I would rather prefer that all the projects have "coordinated" dead lines. At the moment, an Einstein WU is two or three times bigger than a Seti's one, and the dead line gives roughly half the time to do the job !

I've been obliged to put 200 on Einstein to manage it, but it's not my wish

John McLeod VII
John McLeod VII
Moderator
Joined: 10 Nov 04
Posts: 547
Credit: 632255
RAC: 0

RE: I'm not sure this way

Message 13279 in response to message 13278

Quote:

I'm not sure this way of making PC work is logic as long as we have short dead lines. This would mean having a lot of WU going over, and would increase server's work.

I think that we are suffitiently able to choose, in the current way, what we want to do. Having coefs of 950 and 50 means 95% of the workload for one project, which is quite the result wanted, and have the second project in backup in case.

I would rather prefer that all the projects have "coordinated" dead lines. At the moment, an Einstein WU is two or three times bigger than a Seti's one, and the dead line gives roughly half the time to do the job !

I've been obliged to put 200 on Einstein to manage it, but it's not my wish


Upgrade to 4.45. 4.45 will balance resource shares over a longer time while meeting deadlines by processing results in Earliest Deadline First when needed. It will not always keep work from all projects on your host at the same time.

j2satx
j2satx
Joined: 22 Jan 05
Posts: 46
Credit: 1650297
RAC: 0

I like being able to process

I like being able to process WUs for two projects by resource share.

I would rather have the option to change the resource share on each computer.

Heffed
Heffed
Joined: 18 Jan 05
Posts: 257
Credit: 12368
RAC: 0

RE: I would rather have the

Message 13281 in response to message 13280

Quote:
I would rather have the option to change the resource share on each computer.


You can if you set the other computer up as a different venue. Sometimes the venues don't work as expected though.

j2satx
j2satx
Joined: 22 Jan 05
Posts: 46
Credit: 1650297
RAC: 0

RE: RE: I would rather

Message 13282 in response to message 13281

Quote:
Quote:
I would rather have the option to change the resource share on each computer.

You can if you set the other computer up as a different venue. Sometimes the venues don't work as expected though.

I am not aware that you can change "resource share" between venues.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.