Print Checkpoint Write to Log

[alancabler]

Noting checkpoint writes in the log (after delay for write to disk) would assist many donors who try to minimize lost work from client shutdowns and would help minimize checkpoint corruption.
Adding the text at -verb 3 might yield an increase of processing power to the project and negates the need to find similar info in the work directory.
Such a change would come under the heading of "donor friendliness".



___________________________
Please do not mistake my brevity as an indication that I am 7im's sockpuppet.

[Jesse_V]

That would be interesting. Isn't that the core's responsibility? They are the ones that write the checkpoints. I'm not entirely certain, but I thought that most cores write checkpoints every frame (1%) or in the interval specified by the checkpoint frequency, whichever comes first. Some cores may follow this behavior, others may not. Some cores are developed by other research groups outside of Stanford.

[Joe_H]

I'm not entirely certain, but I thought that most cores write checkpoints every frame (1%) or in the interval specified by the checkpoint frequency, whichever comes first. Some cores may follow this behavior, others may not.

Some cores follow both methods in my experience, write a checkpoint at each frame and also at the checkpoint frequency. Not sure if some of the ones I have seen do that are still in use currently.

[bruce]

I'm not entirely certain, but I thought that most cores write checkpoints every frame (1%) or in the interval specified by the checkpoint frequency, whichever comes first. Some cores may follow this behavior, others may not.

Some cores follow both methods in my experience, write a checkpoint at each frame and also at the checkpoint frequency. Not sure if some of the ones I have seen do that are still in use currently.

The GPU cores currently write checkpoints at every frame (1%) and ignore the checkpoint interval. That was fine when frames times were a few minutes long but some of the new projects have much longer run times (depending on your GPU, of course) and checkpoints also become much longer. I've been trying to get this changed in a future version of a GPU core but it hasn't happened yet.

[7im]

Yes, sorry, not a V7 client issue, but a fahcore issue as described. Maybe a mod can move this discussion to an appropriate section.

[P5-133XL]

I'm not sure that it doesn't belong here -- The logs are created by the clients not the cores. Look at a v6 log and look at a v7 log: They really don't look the same and don't include the same info but use the same cores. He's not requesting a change in the behavior of the checkpoints but rather an addition to what gets written to the log at a high verbose level.

PS. I think verbose 3 is too low because there would potentially be lots of checkpoint entries. Something better for verbose 5.

[7im]

Checkpoint functionality is controlled by the fahcore. If the fahcore doesn't produce the data alancabler wants to be written to the log, updating the client/verbosity won't help, so not a client issue.

But since this change would also involve a minor update on the client just to write the new data to the log, it can live here instead of moving to General FAH, IMO.

[alancabler]

Fahcore issue- right! Should have realized that, since the issue isn't client specific.
Thanks for clarification.

[bruce]

Setting 'extra-core-args' to '-verbose' does add that information to the log on some cores. (I have not experimented with every core.)