Hello,

in the advanced view of the BOINC Manager I can open the properties from each task. I would like to see there the estimated time to the next checkpoint.

Background: I run the BOINC Manager, when I do something on my PC. When I'm finnished and want to turn off the Computer I look how long the task had been caculated since the last checkpoint and stop them at a time, that nealy none work is lost by power off the PC. I know the option, to log the checkpoint, but that helps not directly.

There are three types of work units for me in this aspect:

1: WU that checkpoint really often and it's not important, to wait for them.
2: WU with middle runtimes to each checkpoints (~10 minutes). Here I can look in the log and estimate by myself, when the next checkpoint will be reached.
3 - the problem case: WU with long runtimes to each checkpoint (several hours). When there is a WU, my PC calculated 2 hours and there is still no checkpoint reached, it is anoying not to know, if the next checkpoint will be in maybe just 5 more minutes or 2 additional hours.

Would that be possible?

All you can do is set your minimum checkpoint interval in the client via the manager or command line. That simply allows any science application to checkpoint at that interval setting.

It is still totally dependent on whether the science application implements checkpointing, in the first place, and what interval the application developers have coded into the application.

You have no control over when an application checkpoints. Only what is desired if the application is amenable.

You have no control over when an application checkpoints. Only what is desired if the application is amenable.

I don't want to control the checkpoints, I just want to know, when the next one will be.

You have no control over when an application checkpoints. Only what is desired if the application is amenable.

I don't want to control the checkpoints, I just want to know, when the next one will be.

Sorry but the simple answer is “when it wants to”.

As each implementation of each application is different there is no way to predict it from our end.

As each implementation of each application is different there is no way to predict it from our end.

I think it don't have to be predicted. With the display of the percentage, the WU has already been done, there should be already some kind of communication between the BOINC Manager and the project application. Shouldn't it be possible to add there a information for the time to the next checkpoint? The application by itself should be able to estimate it.

I don't think that will work. Some project applications may checkpoint at specific progress %age points, but you still have the time estimation problem - different devices run at different speeds. Other project applications are free to checkpoint at almost any time, and the actual interval is determined by the preferred interval set in BOINC.

I think your best bet is to set <checkpoint_debug> event logging, and monitor the actual behaviour of your preferred project applications on your particular system for a day or two. That should give you an idea of what to expect in the future.

When the checkpoints are at specific checkpoints, than this could be displayed. Than this could be calculated by the current percentage per hour rate - by the manager or by myself. The applications, that checkpoint almost every time could send a flag, that say exact this.

The checkpoint debugging I have already mentioned. Like now I have a work unit, that calculate since 2 hours and have reached not a single checkpoint. In this case i would like to know: Is the Checkpoint in 5 more minutes or 5 more hours? Even percentage could be a indicator. It has calculated ~11,5%. Will the next checkpoint be at 15%, 25%, 50% or maybe don't have it checkpoints at all?

The problem by this watching experience: I have joined 12 projects and some of them with multiple applications. And I don't run my computer 24/7. This is the fact why I want to know the checkpoints; because I want to know, when will be a good time to shut it down. So one or two days will definitly not be enough to figure this out; one or two month would be more realistic.

It is not really possible to remind every single runtime. Yes, I could do a protocol with notice every single application. But "why" should I do that, when it should be also implemented by the application itself? I don't see any technical problems, that makes it absolutly impossible.

Edit: The same application, that reached the first checkpoint now after ~160 minutes, now checkpoint every ~7 minutes. So simply watch the work units and notice the checkpoints seems to be a real science I would really like to avoid.

When this calculation will not come, could there be a option to stop a task at the next checkpoint?

While, in theory, that would be possible it is highly probably that doing so is down to the individual application :-(

Also, some projects tasks do not checkpoint - tasks many only run for seconds to a few minutes or the tasks program have no way to checkpoint.

While, in theory, that would be possible it is highly probably that doing so is down to the individual application :-(

It should be possible to combine it with the option to log the checkpoints; BOINC can see, when a checkpoint is reached.

Checkpoints are determined by the person that wrote the application, and their visibility to BOINC is solely in their hands. The applications that I've looked at work on a hybrid of time and progress, using the BOINC provided "time between checkpoints" as a rough guide, not an absolute figure. Many actually control the write to disk themselves and don't actually tell BOINC that they have done so.

When this calculation will not come, could there be a option to stop a task at the next checkpoint?

BoincTasks (a replacement for Boinc Manager) has this option. It also can show how long a go a task did a checkpoint and how many checkpoints it has made. But it won't show when next is due.

Just google efmer boinctasks.

That is based on the checkpoint interval set in BOINC, not the real interval from within the application - it's just a "guess", not reality.

BoincTasks (a replacement for Boinc Manager) has this option. It also can show how long a go a task did a checkpoint and how many checkpoints it has made. But it won't show when next is due.

If you have checkpoints displayed in BT on the tasks tab, it highlights tasks that have exceeded the checkpoint interval in red.

---

MarkJ

When this calculation will not come, could there be a option to stop a task at the next checkpoint?

BoincTasks (a replacement for Boinc Manager) has this option. It also can show how long a go a task did a checkpoint and how many checkpoints it has made. But it won't show when next is due.

Just google efmer boinctasks.

Thank you for that hint, I will look for that.