Is anyone else experiencing this?
I started running CPDN according to my BOINC window just over 100 hours ago. It told me, having analysed my system, that it would take approximately 870 hours to complete my WU. Now, after having been running it for just over 100 hours, it is telling me that I have over 900 hours left to run. As 100 + 900 is quite a lot greater than 870, I thought I would monitor the system for a little.
Even though the \'Elapsed time\' is incrementing neatly at roughly 1 second per second, the \'Time Remaining\' is not decrementing at the same rate. Task Manager tells me that I am using all 90% of CPU power that I set in settings on BOINC and the vast majority of the memory that is in use (although there is plenty to spare). Is this normal? Why are the counters not incrementing and decrementing equally? All input most welcome!

---

Seb from Deepest, Darkest, Dorset

Hi,

The initial estimate is usually unreliable, since it is based on the \'Boinc benchmarks\' which produce fairly useless figures.

Since the climate model runs at a very stable rate, the best way to estimate how long a model will take to run is to calculate it by dividing the cpu-hours-so-far by the fraction completed so far.

e.g.) 100 hours, 0.1 done (10%) gives us a total of 1,000 hours.

---

I'm a volunteer and my views are my own.
News and Announcements and FAQ

Thanks. That\'s helpful at least in terms of understanding the not-very-good BOINC benchmarks. It does however make me wonder that techies who are clever enough to design and manage BOINC make such bad benchmarking software!

Moving on, it also begs a question: if BOINC were predicting a completed WU from me within x hours, when my result doesn\'t in fact appear after x hours, what will happen to the work that I have done? Will it be ignored or abandoned? Currently, based on the time taken to date, my WU won\'t be finished for another 110 days or so - rather more than three times the total time predicted originally by BOINC. Is this going to be a problem and am I wasting both my time and my CPU time?

Regards,

---

Seb from Deepest, Darkest, Dorset

BOINC was originally developed for SETI, which has short work units, of a few hours each. As more projects have appeared, they have proven to be of a duration lasting from a few minutes to a few days.
All except Climate Prediction models.

There are now, (I think), 70 or so projects, and BOINC has become more and more optimised for these other, shorter projects, simply because they make up the bulk of the projects and users.

One of the reasons for benchmarks, is that the other projects have a deadline for when the results need to be returned, and they REALLY MEAN IT!
SO BOINC needs to be able to estimate how long a work unit will take for each project being run, so that it can decide if another project\'s workunit looks like running past it\'s deadline. If so, it will get worked on next so as to NOT miss it\'s deadline.
And some people run dozens of projects, with each one having, perhaps, a dozen WUs on the go.

This project, on the other hand just uses a very long deadline to keep BOINC happy, and then ignores it anyway.

BOINC WAS/IS written by very smart people, and has become very complex, and very good in situations as above. PROVIDED THAT PEOPLE DON\'T KEEP FIDDLING WITH THE SETTINGS! Which will mean that BOINC has to re-learn things over and over.

It usually takes several WUs from a project for BOINC to \"learn\" how long it takes. It has a few settings to help it remember this, and these can be seen on people\'s Account pages.

And you\'ll find that as your model progresses, BOINC will \"learn\" about it, and the estimates will get more accurate. But the model will nearly be finished by then.
As for your second question, there\'s no such problem. Your model will just keep running until it finishes. Or crashes

Crashes can be caused by:
Unviable model parameters (which is part of the experiment)
Software conflict
Hardware instability
Operator Oops!

---

Backups: Here

OK. That\'s very clear. Thank you. I still find it a little strange that both counters should increment but I suppose c\'est la vie. I equally find it strange that the techies can devise a system which has a \'reliable\' counter in terms of time elapsed, and, another which displays an accurate (you implied) percentage of the work that has been completed and yet a humanoid crudely dividing one by the other is more accurate than their presumably far more sophisticated predictive tool.

Anyway, it now makes me think rather harder about backup however because it would be galling to lose quite so many hours of work through the various possibilities that you list.

Many thanks for your help.

Regards,

---

Seb from Deepest, Darkest, Dorset

The Boinc manager was written in Berkeley and supports many projects (the climate project is based in Oxford). So the Boinc manager doesn\'t know how the climate model behaves in terms of whether or not the %complete figure is accurate. As you say, if the Boinc manager used the %complete figure instead, it\'d be much more accurate for this project, but for other projects different ways of estimating the time might be better.

Backup is a very good idea :-)

There are three different types of climate model, which is why your two PCs have such different time estimates. I\'d recommend browsing through the various README posts for background information - the \'running the model\' one has links to a couple of posts which describe the different types of model available.

---

I'm a volunteer and my views are my own.
News and Announcements and FAQ

Since the climate model runs at a very stable rate, the best way to estimate how long a model will take to run is to calculate it by dividing the cpu-hours-so-far by the fraction completed so far.

e.g.) 100 hours, 0.1 done (10%) gives us a total of 1,000 hours.

I was hoping you weren\'t going to say that. Right now my poor 1.1ghz machine is reporting 91hours of CPU time - Progress 1.01%... so by my calculation that means it will be about 9100 hours to get it all done... and 9100/24 is 379.16 ... well over a year in raw computing time.

However this computer is mainly for my useage... therefore it\'s often only doing computation during the idle time and furthermore I turn it off at night. Is it even worth it to have me continue? or should I wait until I have a newer, more powerful computer before bothering with CPDN?

Hi Idle Computer

Unfortunately your computer is hidden so we can\'t see what type of model you have. You\'ll have to decide for yourself whether to commit your computer for so long. If you do continue, make sure that in your preferences you have \'Keep in memory while suspended\' set to YES, disable the screensaver if you haven\'t already, and try to let the model get just past a checkpoint before turning the computer off. To see the checkpoints in the graphics window, open it then press Z then 8 on the keyboard. Exit from BOINC when the timestep number\'s high. These steps will let the model run as fast as it can in its current home.

If your model runs past its deadline the CPDN server will still accept it.

When a model is going to run for so long it\'s particularly important to make regular backups to avoid the disappointment of losing the model after a crash or other mishap. You can get to the README about how to back up the contents of BOINC folder through my sig.

I ran a 160-year HADCM model for nearly a year on a 1.3GHz computer and it had still only done about 60%. When I then got a new much faster computer I transferred the model to it and let it race to the finish. If you\'ve already had plenty of practice backing up the BOINC folder, the transfer is easy.

Hope these ideas help you decide what to do.

---

Cpdn news

Another alternative is to run a HadSM3 or HadAM3 model instead - these model less time. Again, see the \'README - running the model\', top two posts in the \'information\' section, for more information about which models are available and how to select them.

---

I'm a volunteer and my views are my own.
News and Announcements and FAQ

Hi Idle Computer

Unfortunately your computer is hidden so we can\'t see what type of model you have. You\'ll have to decide for yourself whether to commit your computer for so long.

Although CPDN is currently only running on a single computer, other computers/networks I\'ve worked on in the past means that I usually keep that information hidden. However I can share the details of the current system: AuthenticAMD AMD Athlon(tm) Processor [x86 Family 6 Model 8 Stepping 0] (like I said, 1.1ghz) running XP right now.

Not the fastest computer in the world but on other boinc projects of course (with smaller units) it seems to have managed well.

If you do continue, make sure that in your preferences you have \'Keep in memory while suspended\' set to YES, disable the screensaver if you haven\'t already, and try to let the model get just past a checkpoint before turning the computer off. To see the checkpoints in the graphics window, open it then press Z then 8 on the keyboard. Exit from BOINC when the timestep number\'s high. These steps will let the model run as fast as it can in its current home.

Alright of the steps here I\'ve turned on the Keep in memory while suspended. I\'ll watch my current system usage to see how that affects other programs. If for example Photoshop or my video editing software shows ill effects, I won\'t be able to retain that setting on this computer. As for the other settings, I\'ve never been able to find the graphics option for this, or most other BOINC projects for that matter. I suspect it has to do with the fact that BOINC is installed in service mode. As such I\'m not sure where to find the information on timesteps.

The current model being processed seems to be a UK Met Office HadSM Slab Model 5.06. The CPDN FAQ seems to indicate that models are selected based on the benchmarks, so I pretty much assume my system would be handling the HadSM3 or HadAM3 models which appear to be the least intensive.

Like I said, I\'m just trying to get an idea if it would be worthwhile to run CPDN using a system such as this, or whether it would be better for all parties involved to focus on other BOINC projects with different resource requirements and distribution methods. For myself, CPDN (though not BOINC as it had previously been installed) was just one bullet point of a list of things that could potentially be environmentally savy so I thought I would try it out and contribute a resource percentage of my computer\'s resources here too. Is there some kind of \'minimum requirement\' that should be listed on CPDN pages for the kind of computer that should be involved in the project?

Recommended System Requirements
It\'s a bit old now, as a new type of model was added a month ago. This requires 1.5 Gigs of ram minimum.

See the option: climateprediction.net preferences on your Account page for more info. It\'s also here that you select which type of model you want to process. If no selection is made, they are issued at random; only the hi-res HADAM3 causes a check to be against your computer.
Benchmarks aren\'t used for selection.

There\'s more info in the READMEs, which you can find through the links in my sig.

---

Backups: Here

Idle, HADSM, which is what you say you\'ve got, is the best type of model for this computer.

Forget about disabling the screensaver - with a service install you haven\'t got it anyway.

But if you\'re not for some reason obliged to have a service install, the ordinary installation which gives you the use of the graphics button in BOINC manager would be a better idea. The graphics display would allow you to pick the best moment (just after a checkpoint) to exit from BOINC and turn off the computer. At present you will be stopping the model at random times and losing on average 1½ days of model time on each occasion.

So if and when you upgrade your BOINC version, consider the question of the service install or not.

---

Cpdn news

So if and when you upgrade your BOINC version, consider the question of the service install or not.

Okay thanks for the advice. I\'ll probably do that. I\'ve tended to do service installs because I\'ve been in multi-user environments... but for my own PC I see your point.

But when you upgrade, during the BOINC installation choose not to enable the BOINC screensaver. It slows the crunching of the model, and you can see the graphics whenever you want using the button in BOINC manager.

---

Cpdn news

And in a service install, there is apparently an option to use: Interact with desktop, which will allow for graphics to be seen.

And in a service install, there is apparently an option to use: Interact with desktop, which will allow for graphics to be seen.

Yes, but you also have to use the service control manager (Control Panel --> Administrative Tools, or run services.msc from the Start menu) to change the logon account to SYSTEM, instead of the user account you specified when installing BOINC.

.... As for the other settings, I\'ve never been able to find the graphics option for this, or most other BOINC projects for that matter. I suspect it has to do with the fact that BOINC is installed in service mode. As such I\'m not sure where to find the information on timesteps.

Les and I have posted on how you can get to see graphics with a service install.

Another way of catching the checkpoint, which doesn\'t involve wasting any cycles at all on graphics, is to monitor the timestamps of files in the BOINC folder. The following information is for Coupled (CM3) models, but I guess you could find an equivalent for the SM Slab model you\'re running.

Within the ...\\BOINC\\projects\\climateprediction.net folder, you should find a folder or folders with names corresponding to the name(s) of the model(s) you\'re running. Within each model folder, I see a subfolder called \'dataout\'. Look at that folder in detail view in Windows Explorer, and sort the files on the \'Type\' column.

You should see one or more \"DAY File\" come to the top. Their timestamps should be within the last half hour or so. Periodically, they\'ll all update to match your system clock: that\'s the checkpoint happening.

If you close down the computer when the checkpoint files are at their most recent, you\'ll be running CPDN as efficiently as possible. If there\'s a big difference between the times of the checkpoint files and your system clock, then it might be worth hanging on a bit - there\'ll be another checkpoint along in a few minutes. You\'ll get an idea soon enough of the interval between checkpoints on your particular machine.