[XS]riptide 2008-06-19 12:40:10 | Right. I've heard it before that it is 'risky' to use one install of Muon/DPAD with many nodes runnin off the files. So what is the risk and what is the chances of that risk. How does the risk scale with the number of nodes used? |
Stephen Brooks 2008-06-19 12:51:42 | Auto-save won't work at all, you're best switching it off. The other conflicts could arise when writing/adding to the results.* files, it's possible if two instances try to add a result simultaneously, the whole thing is blanked out. Other than that I don't know of any problems. I guess the risk will scale as the number of nodes squared if it's the file conflict issue. |
[XS]riptide 2008-06-19 14:10:55 | OK. Right. Well I I'm trying it now. Interestingly the auto threading is not as efficient on a remote node, so I simply run two clients to swallow up close to 100% CPU. |
[XS]riptide 2008-06-19 15:55:10 | Well it isn't looking great so far. Very little movement in the results files. The network traffic is 20kBytes up/10hKbytes down per Intel Quad core. |
[XS]riptide 2008-06-19 16:02:13 | And it also led to about 10% CPU usgae of the 'system.exe' process on the host machine. I can only guess that this was due to very aggressive read/write requests across the network? |
Stephen Brooks 2008-06-19 16:14:48 | It should mostly only be reading files during the initialisation of the simulation. |
[XS]riptide 2008-06-19 17:05:20 | Ya. its a good but more. A continous flow.... One I ended the nodes the network flow went to Zero. |