- 1 What might I want to find out before starting?
- 1.1 Is it safe? – See BOINC FAQ: Security
- 1.2 How long will it take?
- 1.3 Why are the work units so big?
- 1.4 Do I need to be on-line all the time while the software runs?
- 1.5 Can I Suspend the experiment?
- 1.6 Can I uninstall the software?
- 2 How-To and other issues with Getting Started
- 3 About the Experiment
- 4 About running Climateprediction.net
- 5 Problems running climateprediction.net
- 5.1 If Models will not download due to lack of hard disk space, how do I fix it?
- 5.2 Why is there 330MB/1GB of data left on my disk?
- 5.3 I cannot complete by the deadline / Are incomplete Work Units useful?
- 5.4 Why have I got a crazy time to completion?
- 5.5 Why won’t it do anything?
- 5.6 Why won’t the graphics display?
- 6 Performance issues
- 7 Questions, Problems and Discussion Forums
What might I want to find out before starting?
Is it safe?
See BOINC FAQ: Security on the BOINC Wiki.
How long will it take?
This depends on how fast your machine is and how often you switch it off and also which Model you are running. The original Slab Model took a 1.4 GHz machine will take about 4 weeks if it is left running all the time. An 800MHz G4 Mac will take almost three months. A 2.8 GHz P4 or Mac G5 machine will take about 3 weeks. We recommend that you do not download the software if you only have your computer switched on for a couple of hours a week. The experiment will run when the screen is switched off.
The sulphur Cycle model was introduced on 26th August 2005. This is approximately 2.8 times longer.
A Transient Coupled Model is due to be launched shortly. This is estimated to take 6.6 times longer than the original slab model (2.3 times longer than the sulphur model). It should be more interesting as it is the first experiment that is meant to be realistic. Previous models have been more about finding out what the model does in response to some fairly extreme forcings.
Why are the work units so big?
How big are the Work Units?
Compared to SETI@Home‘s Work Units, the Models are very big. Even on a fast Pentium 4 it can take around 3 weeks for a Slab Model, 8 weeks for a sulphur cycle model, and 20 weeks for a Transient Coupled Model.
Couldn’t they be made smaller?
SETI@Home timeslices its working units. Climateprediction.net (CPDN) Work Units are a Climate Model that runs for between 45 and 200 model years depending upon which Model is being run. Trying to timeslice a model would make it very inefficient because you have to wait for the results of the first timeslice before the second timeslice can be started. So it doesn’t help and you would have to move around a lot of large (possibly over a hundred MB) data files. Incidentally, when you really get into this project it is nice to see a whole model evolve rather than only part of it.
It is not practical to split a model into different areas because of the huge amount of communication between adjacent cells (3 times per timestep and over 750000 timesteps per model). It isn’t even practical to do this on a network drive let alone connecting only via the internet.
Thats enormous – does that mean I won’t get credit for ages?
No, you will get credit for each Trickle and there are 72 Trickles in a Slab Model, 120 in a sulphur Cycle model and around 2000 in a Transient Coupled Model. So at worst you only have to wait for about 12 hours to trickle on a fast PC running sulphur. The stats updates are currently once per day.
Also you do not have to wait for someone else to validate your work.
In climateprediction.net the Work Units are a complete Climate Model. It stands alone, the project needs lots of slightly different models. The models are further subdivided into Phases, Trickles, and Timesteps.
A Timestep represents a 1/2 hour of model time (not realtime). Every 10,802 timesteps your model will Trickle (report), and there are 24 Trickles in a phase. When you know that a model year is 12 months of 30 days (not 365 days so you do get 30 Feb Dates), you can work out that a Phase is 15 years 1 day of model time (not realtime).
Here’s some rough performance numbers for a reasonably quick P4, using 1 logical cpu:
- Timestep = 2.2 Seconds
- Trickle = 6.6 Hours
- Phase = 6.6 Days
- Model = 19.8 Days
- Timestep = 3.7 Seconds
- Trickle = 11.2 Hours
- Phase = 11.2 Days
- Model = 56 Days
See also Climateprediction.net Size Table
As you can see, for a reasonably quick PC, slab and sulphur models will Trickle (report) about 2 or 4 times a day, and will complete in just under 20 days or 56 days. That is, if it were running 100% of the time.
- Timestep = 2.7 Seconds (but 72 timesteps per day instead of 48)
- Trickle/(Model Month)= 98 minutes
- Model year = 19.5 hours
- Hindcast/Forecast = 65 Days
- Model = 130 days
Irrespectively, you will not see credit in your stats until after the first stats update after your first trickle has been uploaded, then your credit will increase with each stats update where you have uploaded one or more Trickles. Stats updates are currently run only once per day.
More information on Sizes in the Climateprediction.net Size Tables.
What are the implications?
More worrying is that a computation error loses more work. What is the appropriate reaction to this? Complaining is unlikely to be useful as trying to make the Work Unit smaller has been considered and rejected as not practical. A better reaction would be to decide to make a backup from time to time so if you do suffer an error, you can recover without loosing too much work. If you want to do this, see How-To Backup BOINC.
The even longer sulphur Cycle model makes another improvement to address this issue by uploading files at the end of each phase. Thus a computation error in phase 5 still yields useful results for phases 1 to 4 reducing the potential for the science for the entire model to be lost.
Do I need to be on-line all the time while the software runs?
Absolutely not. Once you have installed the software and registered with the project, you are free to disconnect from the internet.
As your computer reaches Trickle points, it will (if set to allow network access) automatically try to “trickle” a small amount of data which is used by the Project to track the progress of your experiment. If no network connection allowed or if unable to connect the Trickle will be stored up until the next successful communication. It is not essential to trickle data. However, an internet connection is required for the final upload of approximately 6 Megabytes of data (which can take about an hour on the slowest modems, for broadband/ADSL it should only require a few minutes) whenever you finish an experiment. Messages will appear in BOINC giving a large amount of detail about how this upload is progressing. The files are split up into five compressed “zip” files, and can be incrementally uploaded (i.e. you can pause/resume or exit/resume and so not have to download the entire file if an initial upload was interrupted).
How is that possible; doesn’t it need to communicate with other computers?
No your computer is doing a complete model on its own. This is why the work units have to be so big – the communication would have to be far too frequent. You cannot realistically do it even on a network drive let alone over an internet connection.
Can I suspend the experiment?
The climateprediction.net Model should not interfere with the normal use of your computer. Nevertheless, you can Suspend all projects via the BOINC Manager’s File Menu. If you just want to Suspend one Project that can be done from the Projects Tab. There is a third possible suspend option on the Work Tab. Note that this one only suspends the Work Unit and if you do this it is quite possible a different Work Unit will (if necessary download and) start.
Alternatively, if you prefer, set your Preferences on the Climateprediction.net website, where you can choose to only run BOINC applications at certain times of the day or only after a period of inactivity etc. BOINC now does “task switching” so running multiple BOINC Powered Projects allows each application a percentage of CPU time which you can set up for each project, so one project does not use up all the CPU time before finishing a Work Unit. You can also set projects to exit when preempted, or to remain in virtual memory (so as not to lose any work).
Can I uninstall the software?
If you wish to stop running climateprediction.net under BOINC, simply “Detach” from the project. In Windows you can do this by right-clicking on climateprediction.net in the “Projects” tab, and then selecting “Detach.” In Linux and OSX open a terminal, navigate to the directory where BOINC is installed and type:
./boinc* -detach http://climateprediction.net.
This should remove all climateprediction.net project files, leaving you free to attach to other BOINC projects. If you want to uninstall BOINC, you should run the “Uninstaller” program from Add/Remove Programs in Windows, for Mac & Linux simply remove your BOINC directory or BOINC executable. We are grateful for any amount of time/energy you are willing to invest in this exciting project. However, please try to finish your experiment once you have downloaded it – do not download the package just to see what it looks like. This constraint is necessary to preserve the scientific validity of the project as partially completed experiments cannot currently be passed on or evaluated.
How-to and other issues with getting started
For “How-to”, see Getting started with Climateprediction.net
If your internet connection uses a web proxy and you use the Windows BOINC graphical client, simply fill in your proxy settings from the “Settings / Proxy Server” menu selection. If you are running the Linux/Mac BOINC client, you will want to set the HTTP_PROXY environment variable to your proxy server. For other options (i.e. authenticated proxy username/password), please see the BOINC Linux or Mac OS X pages.
Note that if you have a firewall you should allow the BOINC client (boincmgr.exe and boinc.exe in Windows; boinc_* in Mac/Linux) to access Internet services on your computer. The client needs to communicate on port 80 (the basic HTTP port used by your web browser). All communication to BOINC servers is initiated by our software as with a web browser – there are no open ports on your computer using this software!
If you are running a firewall make sure that BOINC has full browsing capability on port 80. Climateprediction.net Models have been known to crash when a firewall blocks access to a mirror server if it is configured to restrict BOINC access to named servers.
Proxy server and authentication
For Windows, click on “Setting” in the menu , and then “Proxy Server” to bring up the Proxy server configuration window.
The Command-Line Version of the BOINC Client Software has optional environment variables and command-line options. See the article “BOINC Daemon – Command Line Interface – Command-Line Options” for more details.
About the experiment
What does the climateprediction.net Experiment do?
Details of the various stages of the experimental design of the project are available on the Strategy Page. There are two models which fall into “Experiment 1″ of the strategy. The first is known as a Slab Model or HADSM3. It is a three Phase simulation using a full atmospheric Model but a simplified ocean. The second is known as a sulphur Cycle model as it includes modelling sulphur in several different compound forms. It is a five Phase simulation instead of three and the timesteps take longer as they are more complicated. This model is about 2.8 times longer than the slab model in total.
More information on the different models can be found in Climateprediction.net Models.
- More information on the climateprediction.net science.
About running Climateprediction.net
How many experiments does the software do?
As many as you let it. The software runs a climateprediction.net Model which completes one experiment, and then returns a small set of data from the completed model run to us. It then starts another, different Climateprediction.net Model experiment. It goes around this loop until you choose to change to a different BOINC Project or uninstall BOINC entirely.
At the end of each Phase the output files from the model are processed. They are converted from one format to another. This processing only takes about 5 minutes at the end of each Phase but during this period there is significant disk access and consequently your machine may become sluggish. If this is a problem, simply close down the whole package or just Suspend and restart it at a more convenient time.
At the end of a model, critical features of interest to the scientists are extracted and stored in 5 smaller files totalling 7MB which will be uploaded. The sulphur Cycle model uploads one file at the end of each Phase – 9MB for the first and 2.6MB for each of the other Phases. The data files are then further compressed into a zip format. Around 330MB are left on your computer for a full slab model, 1GB for a completed sulphur Cycle model.
In the Transient Coupled Model, there are uploads every 10 model years. Only 600MB of disk space is needed and the model should delete the files once they are no longer needed.
If models will not download, see Fixing download due to lack of disk space
Running BOINC as a service
- Open a DOS window and change directory to the one where you installed BOINC.
- Install the BOINC service by executing the command boinc_cli.exe -install
- Open the BOINC entry in the services control panel. Click on the Log On tab and make sure it is set as Local System account
- Open file explorer and go to your BOINC directory. Make sure the local system account has full access.
- Exit the BOINC GUI interface and remove it from any startup folder it is in. If you miss out this step the CLI service and the GUI will both start on the next boot. When the GUI starts it will generate a -144 error (shared memory file can’t be created because it already exists) and your current climateprediction.net Model might get trashed.
- Start the BOINC service (either from the services control panel or using the command net start boinc.
System shutdown procedure
If you are running BOINC Client Software on Windows make sure you exit the program before shutting down the system. The BOINC Client Software has no control over how Windows shuts things down, and you can get a corruption of the client_state.xml file which may cause the startup to fail when the system comes back up. The Command-Line Version seems to be safe when it’s run as a service, but running:
net stop boinc
before shutting down will make sure.
Command line arguments
The Command-Line Version of the BOINC Client Software has optional environment variables and command-line options. See the article BOINC Daemon – Command Line Interface – Command-Line Options for more details.
Ready to Report???
This means that the core client is ready to report & will do so when next it contacts the scheduling server. Action to be taken : Leave it alone.
Should I upgrade to BOINC from CP classic?
- The visualization requires a graphics card that has OpenGL support.
- To test your card for GL compliance, download the OpenGL extensions viewer from here.
- Run the standard base install and ensure that your card is 100% compatible.
Problems running Climateprediction.net
If Models will not download due to lack of hard disk space, how do I fix it?
- Note that each Slab model Work Unit requires 750MB of hard disk space. If you have a hyperthreaded processor this means 1.5GB of spare space is required for two virtual processors. The sulphur Cycle model requires 2.7GB of disk space before it finishes, so 5.4GB for a hyperthreaded processor.
- Check your Disk Usage Preferences by going to ‘Your Account’ in the left menu from the website and select ‘View or edit’ General Preferences. Check all three Preferences (Use no more than (MB), Leave at least, and Use no more than (% of total space)) are compatible with providing at least 750MB (2.7GB for Sulphur Cycle) per model.
- If all three Preferences are compatible with providing at least 750MB (2.7GB for Sulphur Cycle) per model, then it could be your Resource Share. Your relative resource allocations control not only how much processing time each of them will get, they also control the percentage of the disk space available to BOINC that each project is allowed to use. This may prevent you from running Climateprediction.net if it is given a low resource allocation (because it uses lots more disk space than other BOINC Powered Projects). A potential solution here is to set the Resource Share for Climateprediction.net very high temporarily until you have downloaded your model.
- If a change is necessary, edit the Preferences and update on the web site. This will reach your computer on the next communication but to speed things up, update from BOINC to get the client to collect the new Preferences. It is now sensible to exit BOINC and restart it. Finally another update should then start the download.
Why is there 330MB/1GB of data left on my disk?
Each Work Unit can leave behind up to 330Mb of data or 1GB for Sulphur Cycle model.
Only a small part (7MB or 20MB) of the data has been uploaded. The rest could still be useful. Obviously the most useful/wanted information has been selected.
The data is left behind because different people will have different policies regarding the Climateprediction.net data.
Some people will want to keep it on their hard disks so they can look at the model with CPView or the advanced visualisation program.
Some people are willing to store data if there is a chance that having it available will save the work being recrunched. The suggestion is that it might prove useful at almost any time in the next 10 years. Most use of the data will probably peak in about 2 years then gradually tail off. Moving it to a CD or DVD is fine – the CP team cannot upload it from your PC without getting in touch with you. Even if some of these CDs/DVD get lost before 10 years is up, at least in some cases it saves the model having to be recrunched.
Others will decide that they are ‘NOT A FREE STORAGE FACILITY’. You are free to decide to delete the data. It is not disasterous, if it later turns out the CP team want the information, the WU can be handed out again to be crunched. Some people will prefer that this should be avoided where possible but it is far better to have people who crunch and delete the data than not have such crunchers in the first place.
Some people have been asked to upload specific runs, so it is not ridiculously unlikely that they will ask for more to be uploaded. Though I don’t think it is a large number so far.
Deciding On a Delete/Archive Policy
If a model has crashed during the first Phase, there is very little point in keeping it. There could be some point in keeping a model that got further as the scientists are interested in finding out how models are crashing. Your policy will depend on how willing you are to save data just in case it may be useful.
Uploads: There is a facility for uploading classic runs. I don’t believe this will take BOINC models. Climateprediction.net don’t have the staff or the storage capacity to cope with people sending in all the information from large numbers of BOINC runs.
There is a “How-To” Save or Delete CPDN Data Files.
I cannot complete by the deadline / Are incomplete Work Units useful?
This question is particularly relevant to sulphur Cycle models which are long and have a short deadline.
The simple answer is that the deadline is not enforced so work will be accepted after the deadline. The deadline is there for a reason, the team want some sulphur Cycle results back by then so that they can start the coupled model hindcast as planned in February 2006. More results will be accepted after this and still be used.
If you are still worried that you will not be able to complete a model, then it is worth noting that incomplete models can still be useful:
An incomplete Phase 1 probably isn’t any use. An incomplete Phase 2, 3, 4 or 5 may at some stage be investigated to find out reasons why it failed. For this, I don’t see any use for a Work Unit that is just abandoned. Also Work Units that crashed due to unstable computers are not very interesting and are a nuisance in that they probably make it harder to find models that have crashed due to their parameters. I don’t know when this might be attempted, so this potential use is somewhat speculative.
The work the scientists want to be able to do first/most involves comparing Phase 2 to Phase 3 also 2 to 4 and 2 to 5. So if you completed Phase 2 but not 3, they couldn’t do such comparisons for your model on your computer but they may still use the Phase 2 for working out the average of the models with the Parameters it has, and compare that average to the average of the Phase 3s that get returned with the same parameters.
Phase 1 returned could be useful for different reasons – it provides flux adjustment measurements.
If you complete 3 phases there is definitely use that can be made of it. They did consider handing out 3 Phase sulphur Cycle runs in order to make them shorter but it just isn’t very efficient needing 9 phases to get the same work as 5.
Why have I got a crazy time to completion?
The Work Units are very large. However sometimes the time to completion displayed is a silly number. This could be due to running Linux version 4.13 or a problem has caused a rewind. In either case, the calculation is finding you have run the model for a long time and have not got very far with the model. The calculation then comes out as very high (which is sensible given the information it is working on).
If you are running Linux Hadsm3 version 4.13
The Linux version of Hadsm3 4.13 has a bug. It works out the percent done for the phase not the model. Consequently at the beginning of phases 2 and 3, BOINC has substantial time on the model and you are shown at an early percent done of the phase. So when BOINC does the calculation in its normal manner, the answer comes out very high. The time to completion does reduce during the phase as the percent done goes up but it cannot be trusted.
If you are sure you are in phase 2 or 3, just ignore the time to completion. Alternatively you can do your own more sensible calculation of
(100 – percent done) * 2592.48 * sec/TS (see your trickles for this) /60 /60 = CPU hours to completion of phase
This bug can also affect the scheduling and causes an extra model to download as you are approaching the end of phase 1. There is little than can be done about this until a revised application fixes the problem.
If you have reached some way into model then it went back to beginning
If this happens you can get crazy high answers for time to completion. The calculation that BOINC does is sensible – it is saying if this return to the beginning keeps happening it is going to take a very long time.
The model is designed to do some rewinds. If it hits a problem it should rewind a day. If it fails again it rewinds a month. If it fails again it rewinds a year but if it fails again it should give up.
If you have got less than a year into a model and it goes back to the beginning it could well be working as designed.
Unfortunately some people hit a problem and end up with their model rewinding to the beginning. This shouldn’t happen and is a serious bug. Sorry no-one has found or fixed this yet. If you have a backup of the BOINC directory, it is probably sensible to restore this. If you haven’t you can continue the model but you will not get credits for redoing the work. You will get Credits for Trickles after the last trickle point you have previously reached. In view of this, you may want to abort the work unit in order to get Credit.
Why won’t it do anything?
This question is a bit general and could be a few things:
- Have you attached to the project? If not see Getting Started.
- Is it struggling to communicate? If so, you may need to change your firewall see Firewall Configuration
- Has it been running but everthing seems to have stopped? Are you running BOINC version 4.45? If so, does this seem to happen with a failed benchmark giving the message Aborting CPU benchmarks, one or more active tasks are still running? If so, exiting BOINC and starting it again may get it going again. However, an upgrade to version 5 is recommended to prevent this recurring.
- Have you set the preference “Do work while computer is in use?” to no? If so, the idle detection may be failing. Consider setting “Do work while computer is in use?” to yes.
Why won’t the graphics display?
There have been issues with ATI graphics cards. A new version 4.13 should fix this.
Another possibility is that your model isn’t running. If it is preempted or paused for benchmarks, this is normal. If it shows as running and you are using BOINC version 4.45, you may want to check to see if you have a Aborting CPU benchmarks, one or more active tasks are still running message. If so, an upgrade to version 5 should prevent this recurring.
Timeslicing between BOINC projects & potential lost work
If you are time slicing make sure you’ve set your preferences to have Leave applications in memory while preempted? set to yes (the BOINC default is no). Climateprediction.net checkpoints every 144 Timesteps, and if you don’t leave it in memory any calculations you’ve done since the previous checkpoint will have to be redone when Climateprediction.net starts up again. If other applications require the memory Climateprediction.net may get swapped out, but that’s a much smaller overhead than you’d get if you were preempted at 143 timesteps!
BOINC GUI display tabs
The Disk and Work tabs do a great deal of communications with the project code. Neither of these should be your normal display tab, even when minimized. A 2 GHz computer can lose 1% of processing keeping these tabs updated. The overhead is much greater for slower computers (a P166 can lose 50% of processing).
For more on performance see BOINC_FAQ: Performance
Questions, problems and discussion forum
You can ask BOINC specific questions on the climateprediction.net Forum. This forum includes a BOINC-style discussion section.
We would like to encourage you to use the forum where participants can share their views, and perhaps provide advice on some of the common problems. Our team members and many knowledgeable participants regularly read the posted messages, so most of your comments will get attention within a few days.
If all else fails, contact the climateprediction.net team. However asking questions on the boards usually works better. If it is serious issue a moderator will usually email someone from the climateprediction.net team.