new function because of new Cr.-system

[Dirk]

Maybe it's possible that you change the function of your nice tool?

It looks like the new Cr.-system don't like rescheduled WUs.

Claggy have a work around if he want to reschedule WUs..
NC subforum : 'But for the love of Gandhi! - Message 1123119' and following..

This work maybe for a low RAC machine..
(and currently I have no idea how to do this.. )

If your nice tool delete 'only?' the related entries automatically, so the S@h scheduler server mark the rescheduled WUs new and the new Cr.-system grant the Cr. well for the rescheduled WUs.
If I understood it correct, because following WUs (not rescheduled) are involved also, they get then also correct Cr.

Your nice tool isn't only nice for to reschedule old re-sent VLAR WUs..
I use an other 'VLAR value', so I reschedule now and in future also. After reading the S@h forum it looks like I'm not the only one.
Maybe the performance of my/the machine/s increase, but the RAC/s decrease.

Maybe an overworked version or maybe addition of this function?

This would be super! *thumbs up*



BTW.
It's maybe possible to do this all automatically (within the tool), without to change always the project settings?

[fred]

After reading the post, I have no clue what the question is.

[Dirk]

If I use your tool, all rescheduled WUs (e.g. GPU -> CPU) are online still marked as they were before (GPU). So the 'new Cr.-system' think they are calculated on GPU (but this guessing is wrong), in real they were calculated on CPU. And calculate & grant wrong Cr.

Claggy's work aroung do, that the S@h scheduler server rename the WUs online, too. And so the Cr.-system calculate & grant the correct Cr. for this and all other following WUs also (later, WUs which aren't rescheduled, they are 'original', are involved later also, because the new Cr.-system add this to the calculation of the Cr.).

[Beyond]

Don't know but here's a somewhat related question.  Does any project besides seti use the new credit system?

[Pepo]

Does any project besides seti use the new credit system?

I think there was some. But please don't quote me for it, I'm not very sure...

[Beyond]

Does any project besides seti use the new credit system?

I think there was some. But please don't quote me for it, I'm not very sure...

Some tried it but AFAIK they all dumped it quickly.  Wondering if anyone but seti still uses it?

[Dirk]

Maybe I explained it not well enough with my poor english..

Here is explained how the new Cr.system work: 'CreditNew'.
I don't understand this..

But if I understood the messages correct in the S@h forum, it look like this..
My GPU calculate WUs.
I reschedule a GPU WU to the CPU.
The CPU WU result was sent to S@h.
But the S@h server think it was calculated on the GPU.
Because of the WU calculation was much slower than expected (after a calculation of a few data on/of the S@h server) it give less Cr.
The S@h server think the GPU is much slower now.
The next GPU WUs get less Cr. than it should.
It need some time to get again correct Cr.

Claggy's work around info the S@h server about the rescheduling (GPU -> CPU).
So the Cr.calculation is correct.

[Beyond]

Maybe I explained it not well enough with my poor english..

Here is explained how the new Cr.system work: 'CreditNew'.
I don't understand this..

But if I understood the messages correct in the S@h forum, it look like this..
My GPU calculate WUs.
I reschedule a GPU WU to the CPU.
The CPU WU result was sent to S@h.
But the S@h server think it was calculated on the GPU.
Because of the WU calculation was much slower than expected (after a calculation of a few data on/of the S@h server) it give less Cr.
The S@h server think the GPU is much slower now.
The next GPU WUs get less Cr. than it should.
It need some time to get again correct Cr.

Claggy's work around info the S@h server about the rescheduling (GPU -> CPU).
So the Cr.calculation is correct.

Maybe the reason almost all projects are rejecting the new credit system is that it is so ponderous and complex that there is little chance of it ever working correctly.  Of course seti has to use it for reasons we won't go into 

[Dirk]

Maybe the reason almost all projects are rejecting the new credit system is that it is so ponderous and complex that there is little chance of it ever working correctly.  Of course seti has to use it for reasons we won't go into 

Maybe you/someone know at least one other BOINC project except S@h (& BETA) which is still using also the 'CreditNew'?
Or at least one BOINC project which used/tested it and canceled it again?

S@h is 'BOINC BETA'.  

Fred, maybe it's possible to 'extend' the BoincRescheduler that the S@h server get info about the rescheduling on the machine - and the WUs are relabeled online also?