Unsent Replicas

Snow Crash
Snow Crash
Joined: 24 Dec 09
Posts: 65
Credit: 100,880,785
RAC: 0
Topic 194704

I noticed that some of the replicas for WUs that I got a couple of days ago not yet been sent out. One of them was sent out and returned with a compute error on 12/31 but the third copy still has not been sent. How does the scheduler handle these? ... seems like they have been forgotten.

I was also wondering what the average turnaround time on this project is.

Thanks,
Steve

--------------------------
- Crunch, Crunch, Crunch -
--------------------------

Holmis
Joined: 4 Jan 05
Posts: 1,118
Credit: 1,055,935,564
RAC: 0

Unsent Replicas

Quote:

I noticed that some of the replicas for WUs that I got a couple of days ago not yet been sent out. One of them was sent out and returned with a compute error on 12/31 but the third copy still has not been sent. How does the scheduler handle these? ... seems like they have been forgotten.

I was also wondering what the average turnaround time on this project is.

Thanks,
Steve

Unsent replicas are quite common here, usually Einstein keeps a work-buffer "ready to send" of about 6-7 days.

Check out the Server status page, there's also a link on the main page near the bottom, the oldest unsent result at the time of this message was created some 6 days 18 h 8 min ago.

/Holmis

Snow Crash
Snow Crash
Joined: 24 Dec 09
Posts: 65
Credit: 100,880,785
RAC: 0

I would have thought that all

I would have thought that all replicas of a WU would be sent out within a short period of time from each other so that there is a better chance that the results would be returned and ready for further processing. Unless of course I am over simplifying the entire process. I guess I could see where you send out replica 1 and only after that is returned do you schedule replica 2 to go out but that is weighted againstyall other replicas waiting so that the *most interesting* preliminary results get their replicas sent out first.

Any stats on average turnaround time for both replicas?

--------------------------
- Crunch, Crunch, Crunch -
--------------------------

Donald A. Tevault
Donald A. Tevault
Joined: 17 Feb 06
Posts: 439
Credit: 73,516,529
RAC: 0

RE: I would have thought

Message 96227 in response to message 96226

Quote:

I would have thought that all replicas of a WU would be sent out within a short period of time from each other so that there is a better chance that the results would be returned and ready for further processing. Unless of course I am over simplifying the entire process. I guess I could see where you send out replica 1 and only after that is returned do you schedule replica 2 to go out but that is weighted againstyall other replicas waiting so that the *most interesting* preliminary results get their replicas sent out first.

Any stats on average turnaround time for both replicas?

A lot of workunits share common data sets, which need to be downloaded to the crunchers' computers. If you have some workunits whose "replicas" haven't yet been sent out, it may be because the scheduler is waiting on another cruncher's computer that already has the required data set. This helps cut down on the amount of downloads that any one cruncher has to endure.

Snow Crash
Snow Crash
Joined: 24 Dec 09
Posts: 65
Credit: 100,880,785
RAC: 0

Thanks, makes sense once it

Thanks, makes sense once it is explained.

--------------------------
- Crunch, Crunch, Crunch -
--------------------------

Comment viewing options

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