3908 (But another "Good" Issue)

[brityank]

I thought I had a problem, until I scanned through this thread. My little molecule showed 5/25 on the Taskbar, and I saw that the Completions were all in steps of 2%. I then noticed interesting timings on the same Folding Project on the two sets I have running:

My older-running Graphical Ver. 5.03
Project: 3908 (Run 1404, Clone 1, Gen 2)

Code: Select all

[01:06:44] Writing local files
[01:06:45] Completed 5000 out of 25000 steps  (20)
[01:28:06] Writing local files
[01:28:06] Completed 5500 out of 25000 steps  (22)
[01:49:40] Writing local files
[01:49:42] Completed 6000 out of 25000 steps  (24)

for 21:34 per step.

and from my newly-running Console Ver. 5.04
Project: 3908 (Run 1106 Clone 2 Gen 1)

Code: Select all

[13:09:43] Writing local files
[13:09:43] Completed 5000 out of 25000 steps  (20)
[13:25:47] Timered checkpoint triggered.
[13:32:51] Writing local files
[13:32:53] Completed 5500 out of 25000 steps  (22)
[13:48:57] Timered checkpoint triggered.
[13:56:26] Writing local files
[13:56:27] Completed 6000 out of 25000 steps  (24)
[14:12:27] Timered checkpoint triggered.

for 23:10 per step.

Seems the Graphical is marginally faster than the Console. Who'd a' thunk it!
(Yeah. I know. Different R/C/G's. But the same Molecule count and composition,
but I don't think the Timered Checkpoint (15Min) is that much a factor.)

[bruce]

Seems the Graphical is marginally faster than the Console. Who'd a' thunk it!
(Yeah. I know. Different R/C/G's. But the same Molecule count and composition,
but I don't think the Timered Checkpoint (15Min) is that much a factor.)

I'm certain it isn't a difference in the client. The times depend on the analysis program FahCore_79.exe which is identical. There are always minor differences between WUs of the same project. (I presume you're running these two clients on the same hardware.)

Unless you know exactly what you're doing, I don't recommend this procedure, but if I suspected that to be true I would move them to the opposite client (keeping the same MachineIDs).

Differences between WUs are a lot more likely than differences between the software.

[brityank]

Seems the Graphical is marginally faster than the Console. Who'd a' thunk it!
(Yeah. I know. Different R/C/G's. But the same Molecule count and composition,
but I don't think the Timered Checkpoint (15Min) is that much a factor.)

I'm certain it isn't a difference in the client. The times depend on the analysis program FahCore_79.exe which is identical. There are always minor differences between WUs of the same project. (I presume you're running these two clients on the same hardware.)

Unless you know exactly what you're doing, I don't recommend this procedure, but if I suspected that to be true I would move them to the opposite client (keeping the same MachineIDs).

Differences between WUs are a lot more likely than differences between the software.

Thanks Bruce. Believe me, I can screw up the code on the phpBB so I don't think I'll attempt swapping the same loads around. Both WUs have already completed and been returned; just thought it to be an interesting observation.

I also have a ~2oC Temp difference between my two CPs on this dual core, and the Graphical runs on the hotter one. Just never paid much attention to the underlying hardware and its relationships until I got into this F@H project; a good learning experience.

Cheers.

[bruce]

I also have a ~2oC Temp difference between my two CPs on this dual core, and the Graphical runs on the hotter one. Just never paid much attention to the underlying hardware and its relationships until I got into this F@H project; a good learning experience.

When you say the Grapical client runs on the hotter CPU-core, how do you know which client and/or which FahCore are running on the designated CPU-core? Are you manually setting affinity or are the FAH processes free to migrate from one CPU-core to the other?

[brityank]

I also have a ~2oC Temp difference between my two CPs on this dual core, and the Graphical runs on the hotter one. Just never paid much attention to the underlying hardware and its relationships until I got into this F@H project; a good learning experience.

When you say the Grapical client runs on the hotter CPU-core, how do you know which client and/or which FahCore are running on the designated CPU-core? Are you manually setting affinity or are the FAH processes free to migrate from one CPU-core to the other?

The Graphical client is always set to Machine ID 1, and there's no "-config" parm to revise it. I have my second Console set as Machine ID 2 to force the second CP of my dual core Intel. Must be working, as my CP0 shows F@H running around 49%, and my Console CP1 running at 46% -- that's because I only allowed it 95% instead of 100% of the available cycles, plus I set it up for 'Low' usage so it doesn't conflict with the Graphical's 'Idle' usage. Maybe that little bit is why it runs cooler?

[bruce]

But the OS does not use MachineID to assign work to a specific CPU-core. That's called setting affinity and FAH doesn't do it.

MachineID is used by the servers to keep track of what WU has been assigned to a specific client but there's no relationship formed with a specific part of your CPU. To demonstrate this, shut down one client (either one) and watch the graphs on TaskManager. If you have a dual core CPU, you'll see one task taking 50% of the processing time and it will be assigned randomly back and forth between the two graphs but the total will stay right around 50%.

[brityank]

But the OS does not use MachineID to assign work to a specific CPU-core. That's called setting affinity and FAH doesn't do it.

MachineID is used by the servers to keep track of what WU has been assigned to a specific client but there's no relationship formed with a specific part of your CPU. To demonstrate this, shut down one client (either one) and watch the graphs on TaskManager. If you have a dual core CPU, you'll see one task taking 50% of the processing time and it will be assigned randomly back and forth between the two graphs but the total will stay right around 50%.

Interesting. My Task Manager shows CP0 with FahCore_78 running ~98%; and CP1 with FahCore_79 running ~96%. Shut down CP1 and the plots go to ~75% on 0 and ~25% on 1, fluctuating by -5/+10; but shutting down CP1 gives roughly equal % on both with little fluctuation. Guess there's more under the covers than I'll understand. Thanks for the info, I appreciate it.

Cheers.