I'm not sure I understand why Taverna would ever want to take all the resources all the time. I thought that if you allocated extra memory, then it would use it only if necessary, for the necessary time required. Surely, Taverna would use this then free up memory by "collecting garbage", thus reducing the memory once again. In effect, you would only have spikes in maximum memory usage - right?
I think I have had the same problem, where you build a workflow and run it. You delete the workflow run (to save memory). You then modify the workflow and re-run it, but this the same amount of memory. It doesn't remove what you have just deleted, but instead doubles up the memory. This is done on a run-per-workflow basis, which might suggest that Taverna isn't freeing up memory. I found the solution to this was to save my workflow, close Taverna, and re-start it from scratch. I'm sure I've already reported this as a bug, but could be wrong. Paul. Stian Soiland-Reyes wrote: > On Mon, Oct 5, 2009 at 08:49, Anja Le Blanc > <[email protected]> wrote: > > >> Yes, that's the reason why I waited for 15 minutes after closing all >> workflows and results. The JVM had all opportunity to run garbage >> collection. >> > > Even if the garbage collection is run at this point, it will never > reduce the memory allocated by the virtual machine. > > One way to avoid the allocation increasing 'unnecessary' is to > manually hint to the garbage collector that it should run while > Taverna is running, say after a workflow is finished or when you close > a workflow run. > > > >> I would not think so. The memory usage was build up over time. I would >> change something in a workflow - let it run - delete the run - change >> the workflow and so on. My impression was that for all the runs new >> memory was allocated but at a delete none was freed. >> > > It is possible to investigate memory 'leaks' inside the virtual > machine, which would reveal if there actually is some memory that is > not freed when you close a workflow run. > > Looking from the outside of the virtual machine you can't really tell > much until you reach the 3 GB 'roof' - if continued runs then don't > stop working garbage collection would kick in and do its job. > > > >> Actually I don't want to reduce the memory used by Taverna, if it needs >> it. All I want is to work for a day without having to restart Taverna. I >> wouldn't have looked at memory if a workflow which had run happily >> before hadn't slowed down dramatically. (I now understand that Taverna >> had reached the 50% memory limit.) >> > > Java's memory allocation is a bit like how we spend our money. Even if > the government gives us a tax break or reduces the VAT, somehow in the > end of the month all of the salary has still magically disappeared. > > If everything is working as it should, then Java and Taverna should > run more smoothly with more memory available, simply because it would > not have to run garbage collection that often - similar to how you > might have gone to the pub slightly more often if your available > income increased slightly. We would in a way just exploit the > capabilities of your machine. > > > However, if you are experiencing that Taverna *stops* working once > you've reached the memory roof, then that is a serious bug that we > would need to investigate. If this is the case, could you possibly > report roughly what kind of workflows you are building and how you are > modifying and running them? (With which data). > > > ------------------------------------------------------------------------------ Come build with us! The BlackBerry® Developer Conference in SF, CA is the only developer event you need to attend this year. Jumpstart your developing skills, take BlackBerry mobile applications to market and stay ahead of the curve. Join us from November 9-12, 2009. Register now! http://p.sf.net/sfu/devconf _______________________________________________ taverna-users mailing list [email protected] [email protected] Web site: http://www.taverna.org.uk Mailing lists: http://www.taverna.org.uk/taverna-mailing-lists/
