pending credit

Zxian
Zxian
Joined: 23 Oct 06
Posts: 40
Credit: 5121474
RAC: 0

RE: I don't really care

Message 88675 in response to message 88672

Quote:

I don't really care about pending credits (or credits anywa), however, the very existence and length of this thread indicates that many people DO care and do not feel comfortable with a large backlog of pending credits (for whatever reasons).

CU
Bikeman


I don't really care that much about a huge backlog of pending credits, but at the same time, wouldn't it make more sense for the project to minimize the number of pending credits? If the same WU is sent to two different machines now, they'll both report back relatively quickly and that WU can be considered finished. If they're sent out months apart from each other, then it's not until the second computer has finished that we know if there's a problem with the results (and therefore can send out the WU to a third host).

Brian Silvers
Brian Silvers
Joined: 26 Aug 05
Posts: 772
Credit: 282700
RAC: 0

RE: BUT if the Project has

Message 88676 in response to message 88674

Quote:

BUT if the Project has a lot of older pc's crunching for it, your idea could eliminate some of them. That is also a balancing act for the Project.

The project had a 14-day deadline in the past. I advocated for an increase to 21 days, but the project felt that 18 was best suited for their needs given the increased processing time with S5R3 and with the level of complaints. The mistake I think you're making is believing that I'm advocating a super-short deadline. You mention 1 day. Others either earlier in this thread or in another mentioned 7 days. I'm mentioning returning it to the original amount of 14 days because the performance of the application has increased, and by more than the amount of the increase in days (as a percentage).

Anyway, just a thought...

mikey
mikey
Joined: 22 Jan 05
Posts: 12712
Credit: 1839116099
RAC: 3624

RE: RE: I don't really

Message 88677 in response to message 88675

Quote:
Quote:

I don't really care about pending credits (or credits anywa), however, the very existence and length of this thread indicates that many people DO care and do not feel comfortable with a large backlog of pending credits (for whatever reasons).

CU
Bikeman


I don't really care that much about a huge backlog of pending credits, but at the same time, wouldn't it make more sense for the project to minimize the number of pending credits? If the same WU is sent to two different machines now, they'll both report back relatively quickly and that WU can be considered finished. If they're sent out months apart from each other, then it's not until the second computer has finished that we know if there's a problem with the results (and therefore can send out the WU to a third host).

A while back over at Seti they instituted a system where any unit that was being resent was sent to a machine that statistically was returning units in 1 day or less. It worked pretty well.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2962505831
RAC: 690773

RE: A while back over at

Message 88678 in response to message 88677

Quote:
A while back over at Seti they instituted a system where any unit that was being resent was sent to a machine that statistically was returning units in 1 day or less. It worked pretty well.


When was that? I haven't seen any sign of it in the two years I've been running Seti BOINC.

mikey
mikey
Joined: 22 Jan 05
Posts: 12712
Credit: 1839116099
RAC: 3624

RE: RE: A while back over

Message 88679 in response to message 88678

Quote:
Quote:
A while back over at Seti they instituted a system where any unit that was being resent was sent to a machine that statistically was returning units in 1 day or less. It worked pretty well.

When was that? I haven't seen any sign of it in the two years I've been running Seti BOINC.

I haven't crunched for Seti in almost 3 years so can't say exactly, but it was working before I left. It was in its infancy and there were bugs to be worked out, but it was working. Rom and Dr. A were even talking of putting it into Boinc itself, not just Seti.

Alinator
Alinator
Joined: 8 May 05
Posts: 927
Credit: 9352143
RAC: 0

RE: RE: RE: A while

Message 88680 in response to message 88679

Quote:
Quote:
Quote:
A while back over at Seti they instituted a system where any unit that was being resent was sent to a machine that statistically was returning units in 1 day or less. It worked pretty well.

When was that? I haven't seen any sign of it in the two years I've been running Seti BOINC.

I haven't crunched for Seti in almost 3 years so can't say exactly, but it was working before I left. It was in its infancy and there were bugs to be worked out, but it was working. Rom and Dr. A were even talking of putting it into Boinc itself, not just Seti.

Hmmm...

Are you sure you aren't thinking about ghost resends? This is enabled here on EAH currently, and they tried it on SAH, but had to disable it due to it bringing the BOINC database and other backend processes to their knees as the amount outstanding work in the field increased.

What I do recall was some discussion of having deadline time out resends getting put into the feeder queue at the top rather that the bottom. IIRC it's a FIFO by default, so currently a resend would get plugged in at the bottom of the list and have to work its way to the top before getting sent.

Alinator

mikey
mikey
Joined: 22 Jan 05
Posts: 12712
Credit: 1839116099
RAC: 3624

RE: RE: RE: RE: A

Message 88681 in response to message 88680

Quote:
Quote:
Quote:
Quote:
A while back over at Seti they instituted a system where any unit that was being resent was sent to a machine that statistically was returning units in 1 day or less. It worked pretty well.

When was that? I haven't seen any sign of it in the two years I've been running Seti BOINC.

I haven't crunched for Seti in almost 3 years so can't say exactly, but it was working before I left. It was in its infancy and there were bugs to be worked out, but it was working. Rom and Dr. A were even talking of putting it into Boinc itself, not just Seti.

Hmmm...

Are you sure you aren't thinking about ghost resends? This is enabled here on EAH currently, and they tried it on SAH, but had to disable it due to it bringing the BOINC database and other backend processes to their knees as the amount outstanding work in the field increased.

What I do recall was some discussion of having deadline time out resends getting put into the feeder queue at the top rather that the bottom. IIRC it's a FIFO by default, so currently a resend would get plugged in at the bottom of the list and have to work its way to the top before getting sent.
Alinator

I don't know what that means, sorry...What I am talking about is since Seti used to use a minimum quorum of 3 returns before credit could be granted, Rom and Dr. A were setting up Seti to re-issue a unit to a faster pc if any one of the first three didn't finish on time or had an error crunching. This made the granting of credits faster and fewer people were waiting FOREVER for their credits. They did this by re-sending the unit to a pc that was returning units within 24 hours normally. I, and many others, got many of these over my time there, the deadline was short, usually 24 to 48 hours, and since I had some fast pc's it was doable.
Einstein could do the same thing, re-issue units with a shorter deadline to machines that are retuning units within 24 to 48 hours, and clear its own cache of pending credits quicker, if they so chose. This would be a change to the current way of doing things, but since they are picking machines that are probably Einstein only, it wouldn't make much difference, except to those waiting for their credits to be granted. People crunching several projects could get into long term debt issues if they sent these units to just anyone though.

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1079
Credit: 341280
RAC: 0

RE: ... Einstein could do

Message 88682 in response to message 88681

Quote:
...
Einstein could do the same thing, re-issue units with a shorter deadline to machines that are retuning units within 24 to 48 hours, and clear its own cache of pending credits quicker, if they so chose...


While now the scheduler has to wait for a machine (for days sometimes) that has the "right" dataset, then it would have to wait for a machine with the right dataset and the desired turnaround time. I don't think that would be really faster.

Gruß,
Gundolf

Computer sind nicht alles im Leben. (Kleiner Scherz)

mikey
mikey
Joined: 22 Jan 05
Posts: 12712
Credit: 1839116099
RAC: 3624

RE: RE: ... Einstein

Message 88683 in response to message 88682

Quote:
Quote:
...
Einstein could do the same thing, re-issue units with a shorter deadline to machines that are retuning units within 24 to 48 hours, and clear its own cache of pending credits quicker, if they so chose...

While now the scheduler has to wait for a machine (for days sometimes) that has the "right" dataset, then it would have to wait for a machine with the right dataset and the desired turnaround time. I don't think that would be really faster.

Gruß, Gundolf


Under that scenario, no it would not be faster!

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.