Backup Projects

[Ghost0210]

I know I should probably email this to the Boinc_Alpha group, but I'm hoping that someone here that knows the Boinc code will know the answer (Boinc Alpha seems a bit busy at the moment anyway), and this has been bugging me for a while

I have a couple of backup projects that I use for times when Seti/Beta have no tasks available or are down etc. But why, when Boinc realises that it can't get any work from either of my main projects does it request a full cache from my backups projects?
It doesn't know how long the main projects are going to be down for so just fills its boots from the backup projects, with no manual intervention from me this ends up with me having a full cache when my main projects come back up so I still can't get any work from them.

Is there a specific reason why Boinc requests this full cache, or does the code that initiates the backup project to get work linked into the normal work fetch coding, so it just looks at the cache I've set and carry on requesting until it is full?
IMO it would make more sense if the backup projects were set to just request 1 sec's worth of work at a time, similar to when you first attach to a project?

[Josef W. Segur]

...
Is there a specific reason why Boinc requests this full cache, or does the code that initiates the backup project to get work linked into the normal work fetch coding, so it just looks at the cache I've set and carry on requesting until it is full?
IMO it would make more sense if the backup projects were set to just request 1 sec's worth of work at a time, similar to when you first attach to a project?

Yes, the "backup project" logic is only tied into the choice of which project to ask for work, beyond that BOINC treats the situation like any other work fetch event.

I agree it would be good to at least offer an option of only requesting 1 second of work for each idle resource. Set and forget systems or those with intermittent connection wouldn't want that enabled, I'm not sure which way the default should be. It wouldn't be difficult to add.
                                                                                    Joe

[Ghost0210]

Thanks Joe,

I guess for the set and forget users and people still on dial up connections for example, having Boinc just ask for 1 sec of work for each resource could cause more issues than it would solve and that the current setting should possibly be kept as a default.
But for more advanced users I definately think that we should have the option to set which way this works, so we can avoid the situations where the backup projects effectively become the main projects for what could be extended periods.

I'll have a think about how to phrase this properly @ Alpha
Ghost

[Ghost0210]

Wow,
Emailed Boinc Dev and DA agreed to doing a Work Fetch for Backup projects which is just enough to fill the idle resources:

I think the right thing to do is to fetch only
enough jobs to fill up the idle devices,
i.e. don't queue extra jobs for backup projects.
I'll check this in.

 

[Josef W. Segur]

Yep, a good idea well presented which is also easily implemented. It's changeset 22148.

I also agree with David that the Set and Forget types won't be using the backup project feature. Intermittent connections as for company computers etc. would be more likely to either have several projects with not much concern about which is active or be restricted to a single project and idleness with power savings would be desired when there's no work for that project.
                                                                                    Joe

[Ghost0210]

I'm just amazed that it was so quick and easy to get a change into Boinc!
If only I could get them through as simply as that at work my life would be so much easier

[Raistmer]

In what BOINC client this feature implemented already?
My BOINC 6.10.58 requested from Einstein ~39 tasks (and downloaded >2GB of supporting data) recently. Einstein's project share setted to zero (0).
Each task takes ~7h to complete, that is, 39 tasks are my full queue - wrong decision for backup project.

[Ghost0210]

In what BOINC client this feature implemented already?
My BOINC 6.10.58 requested from Einstein ~39 tasks (and downloaded >2GB of supporting data) recently. Einstein's project share setted to zero (0).
Each task takes ~7h to complete, that is, 39 tasks are my full queue - wrong decision for backup project.

think it went out in 6.11.6
I'm currently back on 6.10.58 testing something. But I'm 99% sure it was working in 6.11.6

[Fredericx51]

Got extremely much Einstein tasks, too and also a few BIG GPUGrid tasks for FERMI GPU.....(GTX480).
Tried to run 2 Einstein tasks, which worked but not for long, 2nd tasks was retrieved by Einstein and had to
restore GPU setting, running 1 task, at a time.
Over 500 WU's trying to UPLoad and a new batch trying to D'Load, a few hundred at least.

Got some MB WU's on all rig's and Back Up Projects, too.
But I wonder, is this how 'Ghosts' , are "created".....
Would be usefull to have BOINC suspending Network per Project, in stead of setting N.N.T., IMO.

[arkayn]

It seems to work on 6.11.7 as DNETC is only downloading one unit at a time to keep the GPU warm.

[Ghost0210]

It seems to work on 6.11.7 as DNETC is only downloading one unit at a time to keep the GPU warm.

Its the only downside of the new way backup projects work. It will only request 1 task per idle resource at a time. Bit of a issue if you only allow Boinc to connect every few hours

[arkayn]

I keep my computers connected 24/7, so that is not an issue with me.