I am sure this has been suggested, but I'll throw it out anyway.


I would like a setting in the manager to choose how many tasks to assign to a GPU, without having to mess with an app_info.xml.

Most GPU apps are so inefficient, you have to do this app_info.xml nonsense with almost every GPU project.

---

Reno, NV
Team: SETI.USA

I am sure this has been suggested, but I'll throw it out anyway.


I would like a setting in the manager to choose how many tasks to assign to a GPU, without having to mess with an app_info.xml.

Most GPU apps are so inefficient, you have to do this app_info.xml nonsense with almost every GPU project.

I would like a setting in the manager to assign project tasks to a particular compute unit :)

On my wish list is per app DCF for 6.12.x,

i've been running a third party version of 6.10.58 with per app dcf for 6 months now, it just works, somewhat better than Seti's and Collatz's server's scaling the <rsc_fpops_est> per app, no knee jerk reaction after 10 validations!!,

Claggy

Trak #812 raised back on 27 Dec 2008. Still waiting.

It's more an optical issue, but BOINC should only recalculate the remaining time of an application, if the percentage value did change.
If not, it should simply subtract one second per second.

This would prevent the remaining time counting upwards while percentage doesn't change, and would be useful for projects where the finished percentage values change rarely.

And I think, this is not that hard to implement.

Well, actually the remaining time counting upwards is a direct result of switching between applications on the same Task Duration Correction Factor and different <rsc_fpops_est> values. These values are sent to the client by the project, while when they're totally off from what they should be, a lot of weird things can happen, including the counting upwards of the remaining time.

When you see things like this happening, you may want to report that at the project, including which value the <rsc_fpops_est> has (can be found in client_state.xml) and what your TDCF for the project is.

Currently you can only specify a single <count>.5</count> parameter within app_info.xml per app. It is applied across all coprocessors of that type and doesn't allow for different values for different devices.

<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>

Example: Machine has a GTS250 and a GTX460 installed. User wants to run 1 wu at a time on GTS250 but 2 wu at a time on the GTX460.

Provide a mechanism where the user can have different <count> values per coprocessor device.

Raised as Trak #1176

not sure if it has been mentioned yet or not, but I'd like to see a check box that says, "finsih work unit before moving on to another" We can set the amount of time, but each work unit is hard to calculate how long it will actually take ot finish, and i don't like to see a lot of partially finished work units while new ones are started. I'd like to finish the ones started first and then move on to others. No matter how many minutes you set it for now you run the risk of having work units hanging while others start.

In addition I would like to see more control over how resources on my computer are used.

In addition, fix the priority running, I see i have some projects due on teh 24th but it is telling me that a project that is due on teh 26th is running priority. Is there a way to tell the client to run those with deadlines first? and then go to the others.

not sure if it has been mentioned yet or not, but I'd like to see a check box that says, "finsih work unit before moving on to another"

Not going to happen. From Do not stop until it is over: at 98% or more, continue till end of task.:

David (Anderson): This has been discussed and rejected; a job may stay at 99% done for a long time.

In addition I would like to see more control over how resources on my computer are used.

???
You mean you want to decide when what runs for how long and such, instead of BOINC doing that? Why not set the project resource shares to something more your liking?

In addition, fix the priority running

There is nothing to fix here. The first thing BOINC will do is try to get ALL the work on your computer in by the deadline. That's why work further away will at times go into high priority mode, to make sure that that work and all in between will make it by deadline. Mostly seen on systems that have too large a cache to begin with, but as long as BOINC is allowed to continue uninterrupted (and without you deciding what's best for the World), it will learn to ask a little less work next time.

not sure if it has been mentioned yet or not, but I'd like to see a check box that says, "finsih work unit before moving on to another"

Not going to happen. From Do not stop until it is over: at 98% or more, continue till end of task.:

David (Anderson): This has been discussed and rejected; a job may stay at 99% done for a long time.

Why has this been rejected? I cannot find the specific parts of this in this thread, perhaps it was discussed long ago? What is the problem with having the task finish before moving on to another? I don't understant why this cannot be done, it sounds relatively simple to do.

In addition I would like to see more control over how resources on my computer are used.

???
You mean you want to decide when what runs for how long and such, instead of BOINC doing that? Why not set the project resource shares to something more your liking?

This is only part of the problem. Setting resource share for some 20 projects that I run is a pain, I don't want to keep going to each different project website to change these settings. The BOINC client should be a place to integrate all of these things to let us have full control over how it runs. Isn't that part of the point of this software?

I want to have more control over what my processors are doing and how much cpu my pc is using, when I tell it to use more RAM, to be honest I see no change whatsoever. That could be due to the actual projects rather than the client.

In addition, fix the priority running

There is nothing to fix here. The first thing BOINC will do is try to get ALL the work on your computer in by the deadline. That's why work further away will at times go into high priority mode, to make sure that that work and all in between will make it by deadline. Mostly seen on systems that have too large a cache to begin with, but as long as BOINC is allowed to continue uninterrupted (and without you deciding what's best for the World), it will learn to ask a little less work next time.

Ok, it's a little odd, but ok it sort of makes sense.

Why has this been rejected? I cannot find the specific parts of this in this thread, perhaps it was discussed long ago? What is the problem with having the task finish before moving on to another? I don't understant why this cannot be done, it sounds relatively simple to do.

Relatively simple? Nay, not that easily done.
There is work out there that runs quickly, within minutes, from zero to 90-95% and then runs for hours on the remaining 5%. (Usually the protein folding projects, using Autodock).

There's also work out there that runs for more than 100%. Einstein has a couple of applications that run multiple minutes for 100%, can even be paused at 100% when switching for another project's app. Enigma's application will run to 100%, reset its progress and do it again to 100%.

Climate Prediction still has models where the last 2-10% can take days, if not weeks on even a fast CPU. You do not want this one to take over as then your other work will time out.

So as you can see, four applications already that do not adhere to your 'relatively simple' idea.

This idea has been around for about as much time as BOINC has been around. But it's mainly impossible to implement. Not unless all projects go use the same source code for their applications --and what's the fun in that?

Why has this been rejected? I cannot find the specific parts of this in this thread, perhaps it was discussed long ago? What is the problem with having the task finish before moving on to another? I don't understant why this cannot be done, it sounds relatively simple to do.

Relatively simple? Nay, not that easily done.
There is work out there that runs quickly, within minutes, from zero to 90-95% and then runs for hours on the remaining 5%. (Usually the protein folding projects, using Autodock).

There's also work out there that runs for more than 100%. Einstein has a couple of applications that run multiple minutes for 100%, can even be paused at 100% when switching for another project's app. Enigma's application will run to 100%, reset its progress and do it again to 100%.

Climate Prediction still has models where the last 2-10% can take days, if not weeks on even a fast CPU. You do not want this one to take over as then your other work will time out.

So as you can see, four applications already that do not adhere to your 'relatively simple' idea.

This idea has been around for about as much time as BOINC has been around. But it's mainly impossible to implement. Not unless all projects go use the same source code for their applications --and what's the fun in that?

Yeah you are right. I have run all of the projects you mentioned, and I do recall those things happending or the time it takes is actually more than a day even on the machine I am using. So it makes sense, it is unfortunate, but it is what it is.

There should be an option to suspend tasks when CPU or GPU temperature is too high. What a high temperature is should be every user can decide by himself in the options.

That means that for every computer configuration out there, BOINC should be able to read CMOS information under Windows, Macintosh and Linux. Since this will require platform specific application programming interfaces (APIs), this will break compatibility for all other platforms.

This is why BOINC will by default not run work on the CPU and GPU when the system is in use, but only when it's idle. It's also why things like this are best done in 3rd party applications, in this case such as TThrottle.

i have a server that runs boinc, it has 4 hard drives of 2TB each, how can i use them all for boinc (and no, i don't want to make a virtual hard drive that combines the 4 hard drives, just a way to set the boinc data folder to multple locations (i want this because of the new storage system that will be added in the coming year
thx

Could "Last Contact" be added to each host returned with the "project/show_user" RPC?

---

I added to the show_user.php RPC and deployed it on SETI@home.
-- David

---

I would very much like to be able to suspend one project's network usage. One project has been having server problems that result in large files uploading all the way then failing. Over and over and over again. Several people on that project have noted the issue, and we don't know when it will be fixed.

Currently, I have to shut off BOINC's access to the network altogether, which prevents my downloading or uploading other work for other projects while this one is failing, or just keep letting the thing bang its head against the wall and slow my connection on a regular basis.

Supposedly a patch was already submitted to do this, years ago, but it was ignored.

I'm not sure whether this has been requested but I'm just wondering whether it is psossible to add more options to control processor usage. So for example when the computer is idle to be able to set a specific processor usage/time, while having a different set for when the computer is use. I would find this to be pretty useful especially on a laptop when I don't want the laptop getting too hot when I run BOINC and am using it at the same time (say watch something on youtube.

If you're on Windows, then try Fred's TThrottle [stickied thread on these forums]. Just set a temp ceiling and BOINC will be slowed down to not exceed that limit, meaning if your user apps generate more heat, BOINC will be further slowed down. The tool can be set with exceptions, and can control BOINC both for CPU and GPU separately. Try it.

--//--

On multi GPU system is it possible to choose which GPU snooze? It would be great to use one of the installed GPU to something else while the other(s) do the WUs.