Parallel Processing

With ever increasing model sizes, engineers rely more and more on fast solver performance. As CPUs reach their practical limits on speed, increased performance can be realized through parallel processing on multi-CPU computers. NEi Nastran supports parallel processing in the PCGLSS linear solver and Lanczos eigensolver. Typical performance increases with 2 processors are 30-50% for the PCGLSS linear and 15-20% for the PCGLSS Lanczos eigensolver. The table below shows a clear performance improvement when using dual processors on real world models of various element types.

Performance Comparison of NEi Nastran Single vs. Multi-Processing

							Single Processor	Dual Processor
		Model Description	Element Type(s)	Number of Elements	DOF	Solution Type	Solver Time (Seconds)	Solver Time (Seconds)	Performance Ratio (Faster)
1		Bracket	Solids	43,625	242,949	Static	156.8	125.4	1.25
2		Bracket	Solids	136,434	409,302	Static	518.1	392.6	1.32
3		Bracket	Solids	216,617	1,073,466	Static	2951.5	2145.6	1.38
4		Regional Jet	Shells, Beams	52,758	215,808	Static	65.6	49.6	1.32
5		Missile Guidance	Solids, Shells, Beams	56,947	314,274	Static	155.9	121.6	1.29
6		Engine Section	Solids	27,946	167,244	Static	35.6	26.2	1.35
7		Brake Housing	Solids	9,278	56,028	Modal	26.5	19.3	1.37
8		Container	Shells, Beams	19,776	101,724	Modal	65.9	52.1	1.27
9		Engine Section	Solids	27,946	167,244	Modal	92.3	70.9	1.31
10		Equipment Panel	Shells, Beams	16,705	100,404	Modal	40.4	29.2	1.38
11		Hub	Solids	133,571	664,335	Modal	715.9	557.3	1.28