Hello! I have a question about the following message which can be found in f04 file.
It says “When spill is indicated, the model is too large to fit into memory.
The job may run faster by increasing the available memory”.
My questions are
- What is “spill” or “spill group”
- How to manage memory properly
- Is it possible to achieve a noticeable increase in analysis performance with fairly large models by managing memory (for ex. I’m currently working on a 820 000 elements, 5 000 000 DOF model)
- Are there any other Nastran settings worth reconfiguring to achieve performance increase.
I tried to study NX Nastran 8.5 Installation and Operations Guide as the message suggests, but I am russian and my English is not that good unfortunately, so I have a hard time understanding all the special terms.
Thanks in advance!
This is from page 4-14 of the Nastran User's Guide:
In UIM 4157:
• MINIMUM MEMORY REQUIREMENT is an estimate of the open core memory that will allow the decomposition to run, but with heavy spilling to disk.
• MEMORY REQR’D TO AVOID SPILL is the estimate of memory necessary for the decomposition to run in “in core”, without spilling to disk.
Note: These two values represent memory requirement extremes. The amount of memory necessary for optimal CPU performance is somewhere between the two.
• ESTIMATED MAXIMUM FRONT SIZE is a function of the model and affects the memory estimates described above.
o MINIMUM MEMORY REQUIREMENT is a function of the front size.
o MEMORY REQR’D TO AVOID SPILL is a function of the square of the front size.
• NUMBER OF NONZEROES is the size of the input matrix, multiply this value by 8 to estimate the size of the input file in bytes.
• The sum of EST. INTEGER WORDS IN FACTOR and EST. NONZERO TERMS is the size of the output matrix.
Note: Multiply the integer value by 4 and the nonzero value by 8 to estimate the size of the output
file in bytes.
• RANK OF UPDATE is the number of rows that will be simultaneously updated during the
decomposition. You can set by either the “rank” keyword or SYSTEM(205).
stran to terminate after printing UIM 4157.
This can be useful for determining a job’s memory and disk space requirements.
Take a look at our tech tips: http://www.cadimageplm.com/learn/tech-tips
specifically the one titled Accelerating Femap with NX Runtime Performance. In your case, if you have a "big memory" computer, you can certainly take advantage of "spare" memory to improve runtime. For the model size shown, your computer will not be able to take any advantage of "spare" memory if it has only 4GB RAM. If it has 8GB RAM, you may be able to make some mild improvements. If it has 16GB or more, you should be able to make some quite substantial improvements, particularly if the analysis is non-linear, or includes contact. Assuming you are not using the ILP64 bit solver, then your analysis is only using about 500MB RAM. The use of the MEMORY and SMEMORY settings allow you to force spare memory to be used for scratch operations. Your RAM is about 100 times faster than a mechanical disk, and about 5-10 faster than an SSD.
Note also that NX Nastran gives you SMP parallel as standard, so make sure you are using your computer's multiple cores as described in the same tech tip.