I have a setting of "connect every 1.0 days". If I start a new project or if boinc begins new downloads for a "dried out" project, it requests 86400 seconds of new work. That's far above what is really needed (the PC is not "always on", boinc does not get 100% of the cpu, the project has not a "100%-share").

So the boinc client should do the neccessary calculations to ask for the right amount of work or only fetch one WU and wait for the end of the download. If then really another WU is needed, fetch the next one.

Norbert

---

One day is 84600 seconds, so that is what you have asked for, eventually as it gets to know your computer habits it will only ask to fill up with the required ammount.

This link to the Wiki might help you further in the calculations made.
Work Buffer

The Wiki is probably the first place to go when you have questions.

Andy

One day is 84600 seconds, so that is what you have asked for, eventually as it gets to know your computer habits it will only ask to fill up with the required ammount.

The Boinc client already "knows" the project shares and the running fraction. What it asks for is ignoring this knowledge. It never learns to correct this error but later it will ask for e.g. 100 seconds where in reality only 5 seconds are needed. So the consequences are only minimal (when in round robin with work for all projects). I whould call this a bug.

Norbert (the client is a black box for me, but this is what I see ;-)

---

When you detach and reattach later, BOINC will lose the Duration Correction Factor information it has built up. Similarly, it won't have any DCF info when you attach to a new project.

The "connect to network about every x days" setting is a constant. You are asking 24 x 3600 seconds of work = 86,400 seconds of work.
While topping off your work cache, it will happen that BOINC overasks on work. So it will ask for 5 seconds worth of work and get a new work unit in at 5 hours running time. That's because there are no 5 second work units. BOINC rounds up the amount to above. Not below what you asked for.

If you don't want BOINC to ask for 86,400 seconds of work, you put your connect to amount a lot lower. You cannot expect from a piece of software to know what your style of working is when you are telling it to do it one way and then go fight it by shutting down the computer for 23 hours a day.

---

When you detach and reattach later, BOINC will lose the Duration Correction Factor information it has built up. Similarly, it won't have any DCF info when you attach to a new project.

I never did a detach/reattach. It happens everytime a project fetches its first WU (after the first attach or after a time of negativ LTD). Is the DCF really used by the boinc client (I don't think so) or should the server correct the sent WU-size with this value?

The "connect to network about every x days" setting is a constant. You are asking 24 x 3600 seconds of work = 86,400 seconds of work.

What I am asking for is enough work for 24 hours. Thats what the parameter name suggests. If the client does something else, I'd call it a bug.

While topping off your work cache, it will happen that BOINC overasks on work. So it will ask for 5 seconds worth of work and get a new work unit in at 5 hours running time. That's because there are no 5 second work units. BOINC rounds up the amount to above. Not below what you asked for.

Everything is fine if this happens.

If you don't want BOINC to ask for 86,400 seconds of work, you put your connect to amount a lot lower. You cannot expect from a piece of software to know what your style of working is when you are telling it to do it one way and then go fight it by shutting down the computer for 23 hours a day.

I do expect from a piece of software to use the information it has. If I would change my style of working I would expect the SW to be wrong for some time.

Norbert

---

Go to the project you're seeing this at. Log on to your account, view computers, select the computer it's on. Then copy & paste these parts here:

% of time BOINC client is running X %
While BOINC running, % of time work is allowed X %
Average CPU efficiency X
Result duration correction factor X

Plus tell which BOINC client you use and which project it is.

---

Go to the project you're seeing this at. Log on to your account, view computers, select the computer it's on. Then copy & paste these parts here:

% of time BOINC client is running X %
While BOINC running, % of time work is allowed X %
Average CPU efficiency X
Result duration correction factor X

See e.g. this host. You will see, that it reported its last result 8 Feb 2006 20:49:04 UTC and at 21:52:14 UTC its LTD allowed new work and the client wanted the 86400 sec of new work (got only 6 WUs = 12 hours estimated which is too much for 1 day). What you can see there is, that the server seems to have no information about the ressource share (which for Rosetta is 25% at this host) and so the client would have to take this into account.

Plus tell which BOINC client you use and which project it is.

Boinc 5.3.15 at this host but it happend with older versions (and 5.3.17) too.

Norbert

---

What you can see there is, that the server seems to have no information about the ressource share (which for Rosetta is 25% at this host) and so the client would have to take this into account.

That's correct, the server doesn't need to know what resource share you have. It's the BOINC client on your system that needs to know this. It is asking for work. The server is just dumb and giving that work that is asked.

Rosetta units are very varying in time, I see. Not a more 'standard' 4 hour task as some projects have. But ranging from a couple of minutes to over 2 hours.
It's for projects very difficult to release work that really only lasts 1 hour on ALL computers out there. Well, really difficult? Impossible.

Btw, I cannot see the data I requested for, as that's shown to the participant only. Hence why I asked for the information. Clearly say how long your computer is on per day, which project it's doing, at which resource shares. Then we may be able to explain what you see as a bug. (And see if it is a bug).

---

That's correct, the server doesn't need to know what resource share you have. It's the BOINC client on your system that needs to know this.

That's what I said :-) The client should not ask for 86400 seconds of work.

Btw, I cannot see the data I requested for, as that's shown to the participant only. Hence why I asked for the information. Clearly say how long your computer is on per day, which project it's doing, at which resource shares. Then we may be able to explain what you see as a bug. (And see if it is a bug).

I forgot.
% of time BOINC client is running 63.9246 %
While BOINC running, % of time work is allowed 99.441 %
Average CPU efficiency 0.856653
Result duration correction factor 0.482784

Norbert

---

No project out there can give you a task timed exactly for your CPU, unless they have all those CPUs as well, overclocked or not. So you get one standard task to work on.

If you let the science app continue on it, the DCF will take 1% off the estimated time you last had and for the next task start with that time. It'll continue doing so until the estimated time about equals the CPU time (not wall clock time).

This means that it may take quite a while for your BOINC to know of one project what to expect in tasks getting in, asking the right amount of time and thus the correct amount of tasks for you to crunch within the "days between connect" you set.

The problem is: Not all tasks are the same length. Not on any project I have come across so far. Which means that your BOINC has a lot of learning to do if you have multiple projects attached. Even if you only have one project attached.

---

Ageless, your replies don't address my "problems". But in the meantime at the boinc_dev mailinglist a thread was started (named: "work fetch and scheduling issue") that is just about the bug I'm talking of.

Norbert

---