I have connected a single host to two projects. How does BOINC decide how much CPU to give to each one? Is it possible to influence?

in you account for each project, there is a project prefernce. The resorce share is set to 100 by default. This means that 100units of time out of the total (in your case for 2 projects, it will be 200. 100x2) will be allocated to the project, or 50%. If you change the resorce share on one to say 500, then it will get 500 out of 600 total, or about 83%.

Well, now I understand better what that setting means.
Anyway I have set 100 for both, but one is getting a good 20% more.
For sure is not critical, may be the scheduling is approximate

Thank you for the answer

a work unit will not be suspended to work on another until it reaches a checkpoint of the end of the work, even if it puts the project's share way over the desigered amount. This is because each time a WU restarts, it will do so from the last checkpoint, so there wont be any wasted time redoing that was lost since the last checkpoint. And yes, the resorce share is more of a "over time" accuracy instead of "at the moment".

Very clear, thanks. It makes a lot of sense. I was just concerned, as I received a WU that is planned to work for kind of 8 months, and I want to be sure that it doesn't block other projects.

The climate projects will checkpoint ever 2 modle months I believe. However, uppon each checkpoint, a "trickle-up" message is sent on the status of the WU so you can get credit crented as you go along, so you dont have to actually finish a WU to get credit. The climate projects are the only type that will do this.

You got it right, it's the climate. Well, what happens if I have to reboot the machine?
What should I avoid to do to prevent destroying all the work done?

No, it will just restart from the last checkpoint. So yes, you will have *some* work being redone, but it should be more than a few minuets. I have had some WUs crash, but in general, this is one of the first things a project will work on correcting. The climate models still can have problem, but its because of the length of the WU and it depends on why the comp restarted. In general, I haven’t had a significant problem. It is more likely for the WU to end prematurely do to the model running outside its defined bounds than for the project to crash from a restart.

Sometimes a project will cause BOINC to process it more than your other projects. This would be caused by either short deadlines or long checkpoint intervals. However, BOINC will try to give your other project additional time to make up for this when it is able to (after a checkpoint or after it no longer is in deadline trouble). You'll notice this when BOINC starts running a different project for extended time period to correct that 20% difference.

An excellent resource for learning more about how BOINC allocates your computer's time is at the Boinc Wiki.

Thanks for the tip, very interesting Wiki. Just for my curiosity, I will track the climate over a few days. As this machine is working 100% on Boinc, and on two projects, it should be easy to monitor how far it goes from 50% of the elapsed time.

By the way, another funny thing is that the expected remaining time keeps growing (after about 5 days, it has grown from 3516 hours to 3527). I guess that this has to do with the fact that it's an estimation. Still, being the project that long, I hope it will start decreasing sooner or later ;-)