Can anyone on MacNN Team help me to get ClimatePrediction working? I've been running BOINC superbench menubar with great success on OSX 10.3.9 on a G5 dual, but every Climate result is returning a "client error". I've detached from the project a few times, installed the updated fortan libraries released in Sept, and discussed with Mac users in the Climate forums. There's a theory there that dual processor machines may have troubles. Whatever the case, I continue to get client errors. Anyone got a fix?

[Snake_doctor]

Originally Posted by Andrew F

Can anyone on MacNN Team help me to get ClimatePrediction working? I've been running BOINC superbench menubar with great success on OSX 10.3.9 on a G5 dual, but every Climate result is returning a "client error". I've detached from the project a few times, installed the updated fortan libraries released in Sept, and discussed with Mac users in the Climate forums. There's a theory there that dual processor machines may have troubles. Whatever the case, I continue to get client errors. Anyone got a fix?

Yea there was a thread about this on the C@H boards. The app is looking for a directory at the root of the boot drive that is not there. If you create a dummy of the directory the models will run. so you have to create the following series of folders on your boot drive:

opt/ibmcpm/lib

I do not remember if you also have to create a dummy file at the bottom of the list of folders or not but I thik you do. But I had the same problem you are having and creating the folders did the trick. They were supposed to have fixed this in the latest software update. In any case there are instruction for the fix at C@H.

Good luck
Phil

Thanks, Snake.

I've read the various fixes on CPDN. The fortran libraries are in "opt/ibmcmp/lib" -- which says it's at the top level of my hard drive. Is this right, or would it be a different level under my "user"? What's the dummy file?

[Snake_doctor]

Originally Posted by Andrew F

Thanks, Snake.

I've read the various fixes on CPDN. The fortran libraries are in "opt/ibmcmp/lib" -- which says it's at the top level of my hard drive. Is this right, or would it be a different level under my "user"? What's the dummy file?

Andrew,

Try This thread in the Mac CP@H discussion for Macs. It explains the fix in detail. It worked on my system. The Libs you are refering to are for running the Sulpher models. You may or may not get one of those. It is kind of a luck of the draw thing.

Regards
Phil

Snake_dr... I've referred to that particular thread many times. And I'm still confused. Are you saying not to copy these sulphur files? Are there other files? I don't see any other lib files in my "user/library/appsupport/BOINC/projects/climateprediction" folder.

Am I supposed to decompress the .zip files in the project folder after I reattach and these files are downloaded by the project?

As Clay describes in his fix from that thread, are we supposed to copy or move the lib files? Does it matter?

[Snake_doctor]

Originally Posted by Andrew F

Snake_dr... I've referred to that particular thread many times. And I'm still confused. Are you saying not to copy these sulphur files? Are there other files? I don't see any other lib files in my "user/library/appsupport/BOINC/projects/climateprediction" folder.

Am I supposed to decompress the .zip files in the project folder after I reattach and these files are downloaded by the project?

As Clay describes in his fix from that thread, are we supposed to copy or move the lib files? Does it matter?

No you don't have to copy the files from anywhere. The project zip files should unzip by themselves assuming you have stuffit installed on your system. You create the directories (folders) described using the "new folder" command under the file menu in the finder. This is just a set of dummy folders so the system can find this directory structure when it goes looking for it. Eventually it will actually put a few files in there but once you create the folders you can just forget them, the system will do the rest. For some reason the CP@H application is looking for these folders but the system does not create them by itself, so a little manual intervention is required.

Now it is true that the file that is placed at the bottom of this little chain of folders can be anything and a lot of people just copy some convenient file out of the BOINC directory, but you could just create the thing using textedit or simpletext the name and location are what is important. There is nothing in the file so it does not matter. The system just needs to see it there.

But I am sensing that there may be a missing step here someplace. I am assuming you have installed BOINC, registered (created an ID) at the CP@H project site, and are ready to connect to the project, having already received your e-mail with your project id code and the proper URL for the project.

If you have done all that,

Shut down BOINC if you have started it. Then simply open the window for your startup drive, and create the directory, as described in the thread mentioned above, on your start up drive. It is just a chain of folders one inside the other with an empty dummy file in the last folder in the chain. THEN start up the BOINC software, and attach to the project using the information from the e-mail (you can cut and paste it from the mail message to avoid typos). The system should download a model and start running.

Regards
Phil

Looks like I'm getting some project credit now... although my acct credit shows all client errors. Maybe this is trickle?

Looking at my BOINC log it says "Unrecoverable error for result 1feo_100087311_0 (process got signal 10)" on the last few CPDN results.

[Snake_doctor]

Originally Posted by Andrew F

Looks like I'm getting some project credit now... although my acct credit shows all client errors. Maybe this is trickle?

Looking at my BOINC log it says "Unrecoverable error for result 1feo_100087311_0 (process got signal 10)" on the last few CPDN results.

When the model is running, it will sen in a trickle about every 3.6 % of progress. That trickle will get about 95 credits. Be aware that the models frequently crash through no falt of the client, but they will post as client error. You might get 50% through a model only to have it fail and start over on a new one.

So if you have in fact sent in a trickle that would mean that you system is working and has done so for at least 10-20 hours. That is how long it takes on average between trickles. The first trickle takes even longer. The way to tell if the model is running is to either open the BOINC manager window with the GUI version of BOINC, or the message window with the Menu bar version.

In the BOINC manager window you will see the clock running on the CPU time. The percent will move up very slowly (5 min to do .01 percent). In the Menu bar version the model will dump data into the message window as it runs, and you can see it doing that (about once every 1-2 seconds).

If that is not the case then you are still not running.

Regards
Phil

I get this error message my BOINC menubar log...

"2005-11-04 13:03:10 [climateprediction.net] Unrecoverable error for result 1hls_100090187_0 (process got signal 10)"

(which seem to correlate with all my client errors" in my credit.)

Does this mean my client is running incorrectly (but I'm still getting trickle)?

[Snake_doctor]

Originally Posted by Andrew F

I get this error message my BOINC menubar log...

"2005-11-04 13:03:10 [climateprediction.net] Unrecoverable error for result 1hls_100090187_0 (process got signal 10)"

(which seem to correlate with all my client errors" in my credit.)

Does this mean my client is running incorrectly (but I'm still getting trickle)?

I am not certain what the signal 10 error is, but there might be something in the BOINC WIKI about it. It sounds like your models are still blowing up right out of the box. I do not know if you are running other projects or not. Assuming you are not, here is what I would do.

Start over from the begining. You must have at least Mac OS 10.3.9 (or higher) for any of this to work, Mac 10.4.2 is even better. Make certain you have at least the latest version of the free "unstuffit" utility iinstalled on your system before you begin. Let the system drain of any Work Units it now has and don't let it download any more (or if you don't care about a few errors just proceed). Shut down the BOINC application. In the "library" folder on your startup drive, there is a folder "applications support". Open that and in there will be the folder "BOINC Data", drag "BOINC Data" to the trash and empty the trash.

Go to the CP@H site and download the "fortran librarys" for CP@H on the Mac and install them (just in case you get a sulfer model). While you are there login (using the long code from the e-mail you got) go to your account and adjust your preferences to "keep application in memory". This is important for CP@H, we can talk more about why later.

Go to any of the BOINC web sites (Preferably use the CP@H site) and download a fresh copy of the BOINC GUI client (it will give you more control of what is going on). CP@H will run using any of the BOINC versions available there, but eventually you will have to have BOINC 5.2.x so you might as well start with that. Install the new BOINC software. If you do this from the SETI website the package comes as an installer (no zip files or any of that stuff). You can install both the menubar and the GUI versions of BOINC, but you can only run one of them at a time and the version numbers must be the same for both of them. Start the BOINC GUI software. open the "Manager window" (this is available from the menu icon). Select attach to new project under the projects tab in the manager window. Attach to CP@H using your URL and number from the e-mail you recieved.

You will know in a matter of seconds if this worked by looking under the "work: tab in the manager window. If you see the CPU seconds incrementing it is running. If you see "Client error" or "Calculation error" appear unter the status column, then the model has failed.

Assuming that you still have that "opt/ibmcpm/lib" directory chain that you created manually, it should work.

Now just so you know there is no advantage to running the superbench BOINC client for CP@H. That client is designed to give a more accurate benchmark of the Mac performance that the standard BOINC application, and as a result generate higher credit claims for most of the projects. it does not actually improve the performance your system (in lager terms) overall. Since the credtis at CP@H are not based on any claims from the client computer (you always get 95 credits for each trickle) there is no real advantage to using it for CP@H.

If you get the model running you can then shut down BOINC and restart it using the same version number of the BOINC menu bar if you prefer that package. You must be certain the versions of BOINC GUI and BOINC Menu bar are the same if you do that however. If they do not match BOINC may cause the model to crash.

Let us know how it goes. There are only a few other things to try if it doesn't. For some reason some Macs just have a hard time getting this project up and running. But once you get it started it should run ok.

Thanx again, Snake_doctor. I am running multiple clients (6 BOINC projects). I tried leaving the CPDN in memory, and it seemed to bottleneck BOINC (log kept saying "another instance of BOINC is running"). I noticed today on the CPDN site that their dev team has acquired a G5 dual, so I'm going to detach from that project and wait a few weeks. Plenty of other crunchin to do. Also, I'd rather wait until mikkyo has a chance to code the superbench 5.25 before upgrading BOINC.

Also planning to upgrade my Jaguar at home to Tiger in coming weeks so I can quit running SETIclassic and catch up to my work machine.

According to the CPDN website, there is a new client for all three platforms. Anyone tried this version yet?

[Knightrider]

Originally Posted by Andrew F

According to the CPDN website, there is a new client for all three platforms. Anyone tried this version yet?

Just noticed this on the CPDN Forum. When I have dried out Einstien (Number 3 on the RAC) I will switch my quad over to them and see how it runs. I know that they recently aquired a dual G5 2.0, so I am hopeful that things will improve here soon, maybe this is a start.

Tolu
Forum moderator
Joined: Aug 5, 2004
Posts: 112
ID: 3
Credit: 224,470
RAC: 135

Message 17610 - Posted 1 Dec 2005 18:31:15 UTC

version 4.22 of the sulphur cycle is available.
We've improved the performance somewhat.
PS: Note there shotly be a mac release ( before the wknd)
as well as a new Advanced viz update to resolve some of the current quirks.

K.