I was wondering how much credit people have pending.
I have 3,097.56 credits pending, using 1 machine. My RAC is going so slow, has gone down from a high of 400+ to 247.
That's the way machines which return the results in a short time are treated...
Often they have to wait for that ones which return their result late...
I suggested (in the 'wish list') to distribute the work among machines with nearly the same 'Average turnaround time'. That would grant credits faster as the results are delivered in (nearly) the same time.
p.s. my pending credit is normaly about 1400, but I think on weekends more systems deliver results so my 'pending credits' gets lower. Currently it's 1900...
That's the way machines which return the results in a short time are treated...
Often they have to wait for that ones which return their result late...
I suggested (in the 'wish list') to distribute the work among machines with nearly the same 'Average turnaround time'. That would grant credits faster as the results are delivered in (nearly) the same time.
One problem with this method is that it increases the possibility of a WU to succumb to the faults of a particular processor. For example, if a new processor comes out that has a turn around time of 1 hour, but makes frequent mistakes (or even on mistake consistantly) and several people got this new processor. Now, a WU get issued to the new processor. Following the same average turnaround time theory, all copies of this WU would only be issued to this new processor. This processor makes the same error in each of the WU's, therefore the result gets marked as valid and the error gets propagated to the next level. Whereas if the given WU was delivered to a representative sample of processors, this one erronious result would have been marked as invalid and discarded.
I can understand why a person would want to have his WU grouped with the same or faster processors. The last result gets the immediate satisfaction of credit, whereas the faster machines have to wait for us slower machines to finish the same job.
There is work underway to make the Scheduler "smarter". When this makes it into the operational systems we should beging to see improvments in the way things "flow".
Believe me, Einstein@Home is REAL interested in keeping their list of work in progress as small as possible.
Roughly 1100... down from
)
Roughly 1100... down from about 1700 a few days ago.
It's doing wonders for my RAC! :-)
"No, I'm not a scientist... but I did stay at a Holiday Inn Express."
Now down to ~900.
)
Now down to ~900.
Cheers,
PeterV.
I have about 1,350. It has
)
I have about 1,350. It has grown from about 700 three days age. So its still under control
Well its grown to about 1,550 units. So the pending have grown a bit more
Braggers ;P
)
Braggers ;P
Paul
0.00
)
0.00
I've gone from ~500 credits
)
I've gone from ~500 credits pending to ~1400 in a little over 3 days.... Something up with the server or WUs? Why the sudden backlog?
"No, I'm not a scientist... but I did stay at a Holiday Inn Express."
As of right now -- pending is
)
As of right now -- pending is 0.00
If I've lived this long - I gotta be that old!
RE: I was wondering how
)
That's the way machines which return the results in a short time are treated...
Often they have to wait for that ones which return their result late...
I suggested (in the 'wish list') to distribute the work among machines with nearly the same 'Average turnaround time'. That would grant credits faster as the results are delivered in (nearly) the same time.
p.s. my pending credit is normaly about 1400, but I think on weekends more systems deliver results so my 'pending credits' gets lower. Currently it's 1900...
Udo
RE: That's the way
)
One problem with this method is that it increases the possibility of a WU to succumb to the faults of a particular processor. For example, if a new processor comes out that has a turn around time of 1 hour, but makes frequent mistakes (or even on mistake consistantly) and several people got this new processor. Now, a WU get issued to the new processor. Following the same average turnaround time theory, all copies of this WU would only be issued to this new processor. This processor makes the same error in each of the WU's, therefore the result gets marked as valid and the error gets propagated to the next level. Whereas if the given WU was delivered to a representative sample of processors, this one erronious result would have been marked as invalid and discarded.
I can understand why a person would want to have his WU grouped with the same or faster processors. The last result gets the immediate satisfaction of credit, whereas the faster machines have to wait for us slower machines to finish the same job.
Jim
Jim
There is work underway to
)
There is work underway to make the Scheduler "smarter". When this makes it into the operational systems we should beging to see improvments in the way things "flow".
Believe me, Einstein@Home is REAL interested in keeping their list of work in progress as small as possible.