Memory Failed Error in Star CCM+ 4.0
Hi All
I am using Star CCM+ for 3D airfoil simulation. My PC configuration is Core 2 Duo Processor, 4GB DDR3 RAM and working in Windows 7 32 bit version. Now whenever I intend to mesh it shows following error. A request for memory failed. This is typically due to insufficient virtual memory. Command: GenerateVolumeMesh ----------------- ----------------- Recoverability: Non-recoverable error: Server Error Although my allocated virtual memory is 7GB and when this error occurs-more than 2GB of memory remain unused as seen from Task Manager. I am using Surface Remesher+ Prism Layer+ Trimmer meshing model. Anybody knows how to resolve this memory problem? Also can anybody send me typical mesh settings for 3D airfoil simulation? Thank you Shamim |
Two things:
First: you know that Star-CCM+ is already at version 6.02 Second: to use more than 2GB of memory for one process you need 64bit version of CCM+ and therefore also of Windows 7 |
Oolauer
Thank you so much for you post. Actually I don't have 64 bit version of Star CCM now. I have to collect it as well as Windows 7. Isn't there anyway of using 4GB RAM with the 32 bit versions? As far I know 32 bit supports upto 4GB memory. Any more scholar suggestions from anybody? I highly appreciate your sharing of experiences. |
You are nearly right. 32bit Windows supports up to 3.5 GB of RAM but only 2GB per process.
So there is no other serious chance than switching to 64bit |
Olauer
Thank you. You are very right. I am getting just around 3.5GB of memory with the 32 bit version out of 4GB installed. Probably now I have no way but switch to 64 bit version. |
Even when you've switched to a 64bit-version of windows and ccm+, you might experience a bad performance. When meshing takes about 3.5GB of memory, the solver will need a similar amount of memory. And swapping memory to the hard disk is not very efficient.
|
Abdul
Thank you. So typically how much memory is required for 3D airfoil simulation? What could be the possible solution in this circumstances? Should I be in 32bit version instead of switching? Quote:
|
Use more threads/CPUs for the run. Each thread should use less than 2GB of memory.
|
Meshing is seriell. And running the solution parallel requires at least the "head node" to handle the complete model. So just forget 32 bit.
OpenSuse 11.2 will be a good choice and 8GB memory shoud be fine for your start. ... and ccm+ does't like ATI cards :-) |
Without enough cpus/power it's no difference when he use 64bit :D
|
Wrong! If you are not able to run the job at all with a 32bit configuration and within a finite time with 64bit your speedup is very very big.
|
On a Core 2 Duo? Sure that is there a difference between 5 or 6 days? :D
Think about the smiley ;) I know that there is a difference but with that hardware no good job is possible. Work with coarse grid - maybe. But efficient work is not possible. 4GB RAM and 32bit OS. Fastest solution (time from now till solution) is to work with 2 CPUs to be able to use that 4GB RAM. When that isn't enough - sure he needs 64bit. But he also need 64bit OS etc. And maybe also more memory. Installing all needs longer that make that 2 CPU 32bit try-run. More than a try-run it can't be with that hardware. |
Well, I did some wing simulations some time ago. The wing was a piece of the wing, with a length about 4.5m (7.5m semi-span for the whole wing). My mesh had about 35 million cells which took about 20GB memory. Of course, it wasn't ideal, from the current point of view, some tuning would have been necessary.
If you are seriously interested in simulating wings, you can't get over switching to 64bit and invest some money in buying suitable hardware. With your current hardware, you might be able to simulate a slice of the airfoil, but nothing which has to do with 3-dimensional effects of the flow field. Quote:
And Maddin, you're right When running a case in parallel, every thread will only need a part of the memory. But first, shamim still doesn't get a mesh and meshing is a serial process - at least for now. And second, a 32bit operating system can handle a maximum of 4GB per process. Just assuming the solver will need the same amount of memory than the mesher (just a rule of thumb), shamim will not benefit even from running it in parallel due to the maximum amount of memory already being used by one process. JBeilke, STAR-CCM+ doesn't need a headnode suitable to handle the whole model. Otherwise you would never be able to run a Formula 1 car on a cluster. The master process needs slightly more memory than the other computing processes, but not that much more. You just need a big machine for meshing, or a lot of time and a tough hard disk (for swapping) - of course all with a 64bit OS. |
Quote:
He have to split the regions and he must select mesh by region. ;) It only needs more time. With windows you only can access to 2GB memory per thread. To use 4GB RAM you have to use 2 threads (run with 2 CPUs). Quote:
I'd like to use QUDA for my desktop work :( |
Lots of thank to abdul099, Maddin, JBeilke for scholar comments. Now I have got a good idea regarding typical memory requirement.
Here, I have another question, Is it possible to run TWO Core 2 Duo processor and 4GB RAM machine in parallel for STAR CCM? But its currently not possible for me to have new hardwares, so I am considering parallel processing for higher computational power. Your valuable suggestions will be highly appreciated. |
Via LAN or what do you want to do?
Use a smaller hint. With that hardware you can't work good. |
Dear Maddin
I want to do parallel processing via LAN. Probably it will not be very helpful but at this point I have to do it with my available limited resources. If you can please help me at this point. |
Under parallel options you can find all settings you need :)
|
u should have 64 bit win7 with 64 bit STAR CCM+ for such high configuration system. problem is that all the memory don't get utilized in your case. that's why u face problem.
|
That's not right. Sure, a 32bit Windows will use only 3.2GB instead of 4, but even 4Gb is only like a drop in the bucket, compared with the usual ressources most people are using for airfoil simulation.
Therefore the problem is not the operating system. The problem are the available ressources. |
All times are GMT -4. The time now is 16:52. |