LHC@home

I've just started using the VM Remote Console, and I have a number of questions:

I've crunched a lot of 3 CPU WUs, but I don't have any Remote Console data for them. Today I crunched a lot of 2 CPU WUs and one 1 CPU WU. The Event nrs. for the 1 CPU WU were sequential, 1,2,3,4..., and each Event nr. occurred only once.

For the 2 CPU WUs the Event nrs. were e.g. 85,80,81,86,87,82,88,83,89,84,85,86,90,91,87,92,88,89,93,94,95,90,96,91. Each Event nr. occurred twice. Was the jumbling of Event nrs. due merely to processor scheduling factors, or something like that?

Would 3 CPU WUs show Event nrs. 3 times?

The 1 CPU WU had 200 events. What are the events? Were they detected in the ATLAS detector, or are they computational events?

A 2 CPU WU had 102 events, but Event nrs. 99-102 only occurred once, as the WU was ending. What is the significance of that? Was the computation short of completion?

Why would Event nr. 97 take 371.4s the 1st time and 57.02s the second time?

What is being averaged?

Where does the uncertainty (+-) arise?

I know this is a lot. Maybe these have been answered in a thread I couldn't find. I sure would appreciate help.

I've just started using the VM Remote Console, and I have a number of questions:

I've crunched a lot of 3 CPU WUs, but I don't have any Remote Console data for them. Today I crunched a lot of 2 CPU WUs and one 1 CPU WU. The Event nrs. for the 1 CPU WU were sequential, 1,2,3,4..., and each Event nr. occurred only once.

For the 2 CPU WUs the Event nrs. were e.g. 85,80,81,86,87,82,88,83,89,84,85,86,90,91,87,92,88,89,93,94,95,90,96,91. Each Event nr. occurred twice. Was the jumbling of Event nrs. due merely to processor scheduling factors, or something like that?

Each ATLAS task gets a pool of 200 events.
The task then starts as many worker threads as configured in "#cores".
Each thread takes an event from the pool, processes it and logs the success.
As the threads work independent from each other you see #cores sequences combined in 1 log.

Would 3 CPU WUs show Event nrs. 3 times?

Yes.

The 1 CPU WU had 200 events. What are the events? Were they detected in the ATLAS detector, or are they computational events?

IIRC BOINC volunteers process simulated detector collisions (=events) but the scientific app is the same that processes real events at CERN's computing center.

A 2 CPU WU had 102 events, but Event nrs. 99-102 only occurred once, as the WU was ending. What is the significance of that? Was the computation short of completion?

99-102?
Makes no sense as the total wold be 201.
99-100 would make sense and would mean that the last event is currently in progress.
This would end in either 100-100 or 99-101 depending on which thread got the last event.

Why would Event nr. 97 take 371.4s the 1st time and 57.02s the second time?

Event processing can use more than 1200s depending on the input parameters and the speed of your host..
That's what you notice.
Hence on a multicore task an arbitrary thread may process more (or less) events than his colleagues.

What is being averaged?

Where does the uncertainty (+-) arise?

Processing time.

Thanks for the insight, Computezrmie.

Also, if you just want to know the progress of the task, you can checkout the graphics pages ("Show graphics" -> "Your contribution" -> "Your job" -> "More information") to get a simple counter of the number of events processed.

Thanks. I'll try that.