In the General preferences there is the option “Switch between applications every”.
Today you can only choose how many minutes etch applications should run before changing to another applications. I would like Boinc to complete a work unit before starting on a new one, then running multiple projects.

The main advantage with using Boinc I think is that you can take part in multiple projects and chare your computer time between them. With more and more projects using Boinc I will probable participate in more projects. For this to work, Boinc would have to handle charring resources between different project in a different way than today. Today I receive a work unit from every project and Boinc work for 60 minutes and switch to another project. This causes me problems already today to meat the report deadline. I have a PIII 450 224 ram, participating in Seti and LHC. With a Seti work unit running for 24H (14-17H before 4.05) and a giant LHC unit likely running for 60+H (a normal takes 4H) I am not happy right now. This wouldn’t be a problem if:

Boinc completed the work unit before starting on a new one.
Boinc considered all projects before downloaded more work.

This would mean:
That the project would take less space on my HD.
The time from download of work unit to report would decrease.
It would be possible to attach to more projects.
More projects = less likelihood to run out of work.

---

Set your connect to setting to a lower number and you will have fewer WUWUs. then you will not go over the deadline......

---

Semper Eadem
So long Paul, it has been a hell of a ride.

Park your ego's, fire up the computers, Science YES, Credits No.

I already have it set to 0.1
The problem is that Boinc downloads more work then it runs out for a particular project.
Let’s say I participate in 20 projects and there is work available for all of them. Then Boinc would download 20 work units, one for each project.

---

I agree that the CPU scheduler needs more work but always finishing a work unit before moving on to the next will not work. CPDN and its 3 week work units kind of mess that idea up. But I do think that if a work unit is within a few minutes of completing when its hour is up it should be allowed to continue. I have seen a work unit paused 3 seconds before it finished :)

How do you propose this thing work? It goes down the list of projects and asks for 1 work unit from each until it decides it has enough? I suppose something like that would work but it probably wouldn't obey your resource share then. BOINC can't change the fact that your computer can only do x amount of work in y hours. With a single 450 MHz machine you really shouldn't attach to more than 2 projects and obviously CPDN is out of the question :) But then again I guess you are only attached to 2 projects and are still having problems. They really need to make BOINC pay attention to the report deadline. I believe it is completely ignored right now. If a work unit is approaching its deadline, it should get priority over the others.

Hang on for BOINC v4.5 - I think some of these problems might get fixed. And there will be more manual control over work units/projects. You will be able to manually pause work units as well as suspend entire projects.

---

A member of The Knights Who Say NI!
For rankings, history graphs and more, check out:
My BOINC stats site

> I agree that the CPU scheduler needs more work but always finishing a work
> unit before moving on to the next will not work. CPDN and its 3 week work
> units kind of mess that idea up. But I do think that if a work unit is within
> a few minutes of completing when its hour is up it should be allowed to
> continue. I have seen a work unit paused 3 seconds before it finished :)
>
> How do you propose this thing work? It goes down the list of projects and
> asks for 1 work unit from each until it decides it has enough? I suppose
> something like that would work but it probably wouldn't obey your resource
> share then. BOINC can't change the fact that your computer can only do x
> amount of work in y hours. With a single 450 MHz machine you really shouldn't
> attach to more than 2 projects and obviously CPDN is out of the question :)
> But then again I guess you are only attached to 2 projects and are still
> having problems. They really need to make BOINC pay attention to the report
> deadline. I believe it is completely ignored right now. If a work unit is
> approaching its deadline, it should get priority over the others.
>
> Hang on for BOINC v4.5 - I think some of these problems might get fixed. And
> there will be more manual control over work units/projects. You will be able
> to manually pause work units as well as suspend entire projects.
>
As I see it, a 3 week work unit from CPDN isn’t rely a problem (well it might be on my pc, but that is a different problem). Lets say that we run 3 different projects, CPDN 50% Seti 30% LHC 20%. Boinc must have some concept of how long time a work unit takes to complete because today it can calculate how many work units it should download if you only are running a single project. We start Boinc and it downloads a 3 week work unit from CPDN and decides it has enough work for now. 2,5 weeks later it decides it needs more work. It has now spent 420H on CPDN, so it should spend 252H on Seti and 168H on LHC before downloading a new CPDN work unit to obey the resource share. Boinc decides it need 50H of work so it downloads 1 22H Seti units and 7 4H LHC units. Next time it corrects the unbalance by downloading 2 Seti and 2 LHC.

Now this works fine if you only running boinc on one computer. If you use more then one you could let every computer manage its own balance to obey the resource share. This is practically how it works today. Or you could set up some mechanism to balance between the computers if you think it is worth the hassle. Ether way it balances out and come closer and closer to the set resource share over time.

---

As far as I know there have been no (or minimal) improvements in the CPU scheduler in CCv4.52(4.53). There are definatly known problems that will be worked on sometime though, there are 2 or 3 seperate taskbase enteries. It is planned to only allow switches at checkpoints so that work is not lost when a project has a long period between checkpoints. It is also planned to take deadlines into consideration.

Personally I like the quicker project changes with the CPU scheduler, however I miss the ability to attach to as many projects as you want from the older work scheduler policy. I made a proposal that has been incorparated into a taskbase entry to buffer the CPU scheduler requests through a long term scheduler. The long term scheduler would work similarly to the old style scheduler and keep the short term scheduler from getting too much work. So computers can be attached to an unlimited number of projects and still make the deadlines.

---

BOINC WIKI

BOINCing since 2002/12/8

Hi

I downloaded the BOINC client yesterday and attached the seti@home and the climate prediction projects to the client. I have 3 seti WUs and 1 climate prediction WU - the strange thing is that one seti work unit runs all the time and every sixty minutes the client switches between processing the second seti WU and the climate prediction WU. This means that for half the day the climate prediction WU is completely ignored which is crazy considering how long the thing would take to process under ordinary circumstances.

Actually while I was typing this I thought of setting the switch interval to zero and that 'appears' to have solved it. Are there any problems I should know about related to setting the switch interval to zero?

---

> Hi
>
> I downloaded the BOINC client yesterday and attached the seti@home and the
> climate prediction projects to the client. I have 3 seti WUs and 1 climate
> prediction WU - the strange thing is that one seti work unit runs all the time
> and every sixty minutes the client switches between processing the second seti
> WU and the climate prediction WU. This means that for half the day the climate
> prediction WU is completely ignored which is crazy considering how long the
> thing would take to process under ordinary circumstances.
>
> Actually while I was typing this I thought of setting the switch interval to
> zero and that 'appears' to have solved it. Are there any problems I should
> know about related to setting the switch interval to zero?
>
>
The switch interval should be set to something larger than the maximum checkpoint time of all projects you are attached to. CPDN checkpoints every 144 timestamps and will be the one you need to base this on. Figure out how long it takes CPDN to checkpoint then set it appropriately.

---

BOINC WIKI

BOINCing since 2002/12/8