Plotting Grid/ Elapsed & CPU time...

[Vesh]

2 random questions I have;

What is the difference in Elapsed time and CPU time which is displayed in the .rpt file once an analysis is finished? They are often very different.

What is the plotting grid option under the Output tab in the Static analysis definition dialog box? I have always been taught to change this to 7.

Cheers...

 

[mch]

Vesh,


Elapsed time = total analysis time, including time taken to write files, swap etc.


CPU time = time spent by CPU analysing problem.


CPU time is always less than elapsed time unless you've got multi-processors (including dual core) where the CPU time is the total combined time for all processors, which can be greater than the elapsed time.


Plotting grid specifies the number of results points sampled by the results post-processor. A grid of 4 (the default) gives 25 results points per element face, a grid of 7 gives 64. The formula is n = (p+1)^2, where n = number of results per face and p = plotting grid number. Only increase the plotting grid number if your model has relatively few elements, say less than 2000, otherwise 4 is good. For very large models (20,000 elementsor more)the plotting grid can be reduced to save disk space and improve results rendering time.


Hope this helps,


Rod

 

[Vesh]

mch,

Thanks for your response.

Proves its better to be "taught" something, rather then "told" something.

Uni texts always "told" us to change plotting grid to 7, (presumably because of small models). Now after being taught from above and doing larger models, it looks like I can decrease analysis time by reducing this plotting grid.

Cheers, very helpful.

 

[jabbadeus]

Great explanation, Rod.


If you are seeing big differences between elapsed (wall clock) and CPU (processor) time, it may be an indication that you want to bump up the amount of RAM allocated to the solver. By big, I mean you're seeing wall clock more than 6-10 times CPU time. You are probably spending a lot of time writing to virtual RAM.The default is 128 Mb. You should set this to half of available RAM, or go into the Task Manager and see how much RAM isn't being used by the computer and set the RAM allocation to that amount (or slightly less).


And increasing the plotting grid to the max (10) can cause the solver not to complete depending on the model size and computer resources. Like Rod says, 4 is usually good for the kinds of results most people are usually interested in.

 

[Vesh]

My difference in time isnt as much as 6-10 times. Usually the CPU time is around 70-95% of Elapsed time. I do have Dual Core and allocate 1 - 1.3GB for analysis's.

I will start changing the plotting grid to 4 if others say it is sufficient.

 

[mch]

Jabbadeus,


You must be careful not to allocate too much RAM, since the figure you specify is only for holding the stiffeness and other matricies. Mechanica needs RAM for other tasks, particularly I/O, and of course the O/S needs RAM as well. I have seen the effect of applying too much RAM is to make the run slower. I would not recommend going above 700-800 Mb no matter how much physical RAM you have installed, even on 64-bit machines.


If you have dual-core and/or multi-processors, set the environment variable MECH_NUM_PROCESSORS = -1. This will make Mechanica use all available processors. You can force mechanica to use the number of cores you want by replacing the -1 with the number, e.g. 1 = only single processor, 2 = two processors etc. Don't put in a number bigger than the actual number of processors you have!


It is very common on RISC based processors (including AMD - the FPU architecture is derived from the DEC alpha) to see CPU times much, much less than the elapsed time. This is the problem with these type of processors, internal I/O is the bottleneck. However, the elapsed time is usually faster than the CISC rival anyway.
Rod


Next Mechanica training course scheduled for 30-Oct-06, see the web-site for details.

 

[gbon]

You must be careful not to allocate too much RAM, since the figure you specify is only for holding the stiffeness and other matricies. Mechanica needs RAM for other tasks, particularly I/O, and of course the O/S needs RAM as well. I have seen the effect of applying too much RAM is to make the run slower. I would not recommend going above 700-800 Mb no matter how much physical RAM you have installed, even on 64-bit machines.

Very interresting, Mch,


So, do you think having 4 GB of Ram instead of 2 GBcan decrease computing time of Mechanica on Amd 64x2 plattform ?


Of course, in this case, 64 bits OS should be use.


More generaly, is there any hardware settings Mechanica comparison available anywhere ?


Thanks

 

[gbon]

If you have dual-core and/or multi-processors, set the environment variable MECH_NUM_PROCESSORS = -1. This will make Mechanica use all available processors. You can force mechanica to use the number of cores you want by replacing the -1 with the number, e.g. 1 = only single processor, 2 = two processors etc. Don't put in a number bigger than the actual number of processors you have!

Mch,


I've tried mech_num_processors = 1 or 2 and it's exacltly the same for my opteron dual core.


I've tried a solid model with single pass.


Only one processor is used for autogem and both cores are used for the two passes.


Do you know why ?


Thanks