- Joined
- Jan 9, 2006
If you don't have more than 1/2 GB of memory per core, then you don't want to run Rosetta with it's default preferences. It will start using your HD swap file, and then your whole system performance goes into the ****ter.
This was the fix for Tanker (the 64 core system), after Rosetta started using ALL the memory (32 GB), to start 64 work units, AND 600+MB of the swap file on the HD! It is best to do this BEFORE you connect to the project and start crunching a work unit.
This will limit Rosetta's memory use to about 88%, and almost remove the use of the swap file, as well. If you have an 8 core system with 4GB, or a 12 core system with 6GB, this should work for you, since you have the same ratio of RAM to cores as Tanker has.
If you are in simple view, click on the "advanced" button, then
1) Click on the advanced pull down menu >> preferences >> disk usage and memory.
Set the percent of memory BOINC should use to 85%.
Set the amount of swap file it can use to 0.00%
Click on OK to save it.
2) Click on advanced again >> read the preferences file.
If you have already started your work units working, and they are already using more memory and the swap file, then you'll get a few work units that will fail with "computation error". You'll also get errors -- indeed I got a dozen of them with Tanker when I did this, including just having the BOINC manager fail to respond to anything, etc.
Finally, you'll get a window saying "Communicating with BOINC client, please wait...", and it just sits there.
Finally (in about an hour), it started working again, but without the BOINC manager (the GUI) working yet. Only the cores were working - you can see that via the cores (nearly) all at 100%, in System Monitor).
I restarted the BOINC manager, and all was well - memory used was down to 88%, and swap file used was down to 0.1%.
Out of 64 cores and work units being originally processed, 2 failed with "computation error", only. About 60 work units now run at the same time, and there are a few that say "waiting for mem...".
All 64 cores are not working - only 60 of them, but that is a good deal to get rid of nearly all the use of the swap file.
All the errors and b.s. could have been avoided, if I had set these preferences BEFORE I started crunching, but I had no idea the work units would want to use so much memory, so that was not done.
Note that the preferences file is for the local system ONLY. It is NOT for all your PC's running BOINC. If you want to set it for all your PC's running BOINC, you need to change the preferences file that BOINC keeps on the web. You get there by going to the "Projects" page, and clicking on the "Your Preferences" button on the left hand side of that page.
Remember, after you make changes to the file, you have to go back in Advanced, and tell it to read the file you just changed. At least, that's the way I interpret having that option in the list of choices.
This was the fix for Tanker (the 64 core system), after Rosetta started using ALL the memory (32 GB), to start 64 work units, AND 600+MB of the swap file on the HD! It is best to do this BEFORE you connect to the project and start crunching a work unit.
This will limit Rosetta's memory use to about 88%, and almost remove the use of the swap file, as well. If you have an 8 core system with 4GB, or a 12 core system with 6GB, this should work for you, since you have the same ratio of RAM to cores as Tanker has.
If you are in simple view, click on the "advanced" button, then
1) Click on the advanced pull down menu >> preferences >> disk usage and memory.
Set the percent of memory BOINC should use to 85%.
Set the amount of swap file it can use to 0.00%
Click on OK to save it.
2) Click on advanced again >> read the preferences file.
If you have already started your work units working, and they are already using more memory and the swap file, then you'll get a few work units that will fail with "computation error". You'll also get errors -- indeed I got a dozen of them with Tanker when I did this, including just having the BOINC manager fail to respond to anything, etc.
Finally, you'll get a window saying "Communicating with BOINC client, please wait...", and it just sits there.
Finally (in about an hour), it started working again, but without the BOINC manager (the GUI) working yet. Only the cores were working - you can see that via the cores (nearly) all at 100%, in System Monitor).
I restarted the BOINC manager, and all was well - memory used was down to 88%, and swap file used was down to 0.1%.
Out of 64 cores and work units being originally processed, 2 failed with "computation error", only. About 60 work units now run at the same time, and there are a few that say "waiting for mem...".
All 64 cores are not working - only 60 of them, but that is a good deal to get rid of nearly all the use of the swap file.
All the errors and b.s. could have been avoided, if I had set these preferences BEFORE I started crunching, but I had no idea the work units would want to use so much memory, so that was not done.
Note that the preferences file is for the local system ONLY. It is NOT for all your PC's running BOINC. If you want to set it for all your PC's running BOINC, you need to change the preferences file that BOINC keeps on the web. You get there by going to the "Projects" page, and clicking on the "Your Preferences" button on the left hand side of that page.
Remember, after you make changes to the file, you have to go back in Advanced, and tell it to read the file you just changed. At least, that's the way I interpret having that option in the list of choices.
