Page 1 of 1
System Requirements
Posted: Thu Dec 18, 2008 11:23 am
by reill111
I was wondering what kind of configuration/system setup people are using. Processor, hardware, etc.
It takes me hours(2-4 hrs usually) to run one pass of CMC (and sometimes RRA). I can't imagine this is the case for everyone out there. Are there system requirements/configurations that the OpenSim team recommends?
(And I know someone will ask, so let me say that using even outrageously large actuator values does not make this go any faster)
Thanks for the suggestions
RE: System Requirements
Posted: Thu Dec 18, 2008 3:29 pm
by danielleib
Hi Kathleen,
I've been running it on a box with 2.66ghz Xeons (Core architecture) and 8GB of 533mhz RAM, though OpenSim is 32 bit so it'll only address up to around a little over a gig before things start going awry. Since the program isn't multi-threaded more cores won't add anything really; I'd think a box with an Intel E6xxx or higher and 2GB of quick RAM (CL5 or lower at 800mhz) ought to run it pretty quickly. This is with XP Pro of course; with Vista you'd probably want another gig or so to avoid swapping things out of RAM due to the Superfetch feature.
When you mention taking a couple hours to run a CMC or RRA pass, how many data points are we talking about? It takes me about 45 minutes to run 5 full revolutions of a cycling task collected at 100Hz through RRA.
I've also noticed that the application will often leak memory if you run multiple simulations in a row, at least in my config. I've found it best to close the program after RRA and re-open it to run CMC when large data sets will be generated. It's possible you might be experiencing the same behavior when it takes a few hours to run the simulation. If you're not familiar with it, you can track this using Task Manager and looking at the mem usage column in the processes tab.
The plot function, however, is an entirely different beast. I've had it take over an hour to plot 16 muscle lengths over the same crank cycles. This was a few versions back, though, and might have been sped up. I also understand that an awful lot of math is being done with this data output.
Have a good one,
Dan
RE: System Requirements
Posted: Fri Dec 19, 2008 12:12 pm
by samner
Hi Kathleen & Dan,
Kathleen, my simulations of a single gait cycle of running usually take about 30-45 minutes in CMC. I think RRA is slightly quicker.
One other suggestion to significantly speed up your analyses, like RRA & CMC, is to run them from the command line. All you need to do is have the proper settings file.
To get to the command line, go to the Start menu, select Run..., type in cmd, and hit Enter. Then, change the directory to the location of your settings file by type cd "C:\InsertYourDirectoryHere" (the quotes are needed it the file path has spaces).
Then, for CMC, just type cmc -S "MySettingsFileName.xml" (again quotes are needed only for spaces). The command line functions are also explained in the User's Guide.
One other IMPORTANT caveat, is if you save your settings file from the GUI, double check all of the files specified in the XML file are correct (it is easiest to save all the files associated with a settings file in the same folder), and make sure that a model file is specified (the GUI will leave it blank).
Doing this will also avoid the "memory leak" that Dan mentioned.
Dan, what version of OpenSim are you running?
Hope this helps,
Sam
RE: System Requirements
Posted: Fri Dec 19, 2008 12:56 pm
by danielleib
Hi Sam,
I'm on 1.5.5, latest version available for typical download.
Does that mean that the leak is a graphical thing then? I'm more than happy to run from the command line; next time I work with it I'll use that and report back.
Have a good one,
Dan
RE: System Requirements
Posted: Fri Dec 19, 2008 1:04 pm
by aymanh
Hi All,
I just wanted to add a couple comments/questions.
There's a size/complexity question to Kathleen, models with different complexity (# joints, types of joints, muscle wrapping, .. since the model is integrated) as well as the size of the data set can make CMC, RRA, .. very expensive computationally. Having a faster machine can speed things up just a little but beefed up machines with quad-core may not help much as of now (since we don't use multiple cores), and in the models I saw memory bounds was not an issue on machines with 1G-2G of RAM.
To compare apples with apples, it would be useful if you give us an idea how long does it take to run the gait2354_simbody model from Tutorial 3 on your machine with no other CPU hungry tasks running at the same time. I'd also be interested to know how big is your model and what features of OpenSim does it use.
2. It makes perfect sense to run RRA, CMC from the command line simply because there's no reason to use the GUI for non-interactive tasks. I'd suggest though that you start things once from the GUI to make sure you got the setup correct before switching to the command line
Best regards,
Ayman
RE: System Requirements
Posted: Fri Dec 19, 2008 1:09 pm
by aymanh
Hi Dan,
It is misleading to look at the memory meter on windows since the application is written in Java and the Java runtime decides on its own when to perform garbage collection to free unused memory.
You may not see the memory usage go down but you'll see that after you close a model, loading another one costs very little memory. If you run from the command line this issue does not exist and memory usage will be just what you're actually using.
If you run into scenarios though where you suspect a leak either from the GUI or command line (where memory usage keeps growing monotonically) please let me know.
Best,
-Ayman
RE: System Requirements
Posted: Tue Dec 30, 2008 1:10 pm
by danielleib
Hi Dr. Habib,
This is actually something that I saw quite frequently when running larger simulations from the GUI, but it is likely an issue with the XP x64 version after running a quick test; I ran a few larger simulations on a 32 bit machine running XP we have in the lab and didn't encounter the problem. Wish I'd realized this a few months ago, would have saved myself a fair bit of time
Thanks,
Dan