MSC.Nastran V2001 Distributed Memory Parallel Test Results

This page contains the results of several jobs running on different hardware platforms. The times listed in each table are elapsed time. The plots show scaling of the Slowest serial time / Each machine. Each job was run by hardware partners. For best price performance results based on Linux please contact Brad Kindorf

Below is a summary of the test problems.

Name	Ndof	Description	SOL	MEM	SCR Disk Used	Total I/O	Comments
LG0QDF0	31,125	Cube w/ interior	108	100 Mb	0.6 Gb	700 Gb	76 Frequency Increments
XL0EMF0	654,560	Car Body	111	400 Mb	11 Gb	328 Gb	448 + 34 Roots
XL0RST0	739,815	Engine	101	320 Mb	4 Gb	18 Gb
XL0TDF1	525,027	Car Body	108	450 Mb	5 Gb	209 Gb	32 Frequency Increments
XX0CMD0	1,584,622	Car Body	103	800 Mb	43 Gb	2400 Gb	1073 Roots
XX0CMD1	1,584,622	Car Body	103	800 Mb	43 Gb	2400 Gb	1073 Roots (FDMODES)
XX0CMD2	1,584,622	Car Body	103	800 Mb	43 Gb	2400 Gb	1073 Roots (ACMS)
XX0DMD0	1,920,855	Engine	103	800 Mb	25 Gb	555 Gb	17 Roots
XX0DMD1	1,920,855	Engine	103	800 Mb	25 Gb	555 Gb	17 Roots (w/ FDMODES)

Below is a description of the hardware used. Detailed computer, disk, network and OS configuration is listed the bottom of this page.

Vendor	Hardware	O/S
Compaq	AlphaServer GS160 EV68 1 GHz	UNIX 5.1
Compaq	AlphaServer SC ES40 833 MHz	UNIX 5.1
Cray	SV1ex 500MHz	UNICOS 10.0.0lf
HP	HP SuperDome 750 MHz PA8700	HP-UX B.11.11
IBM	IBM p690 Turbo 1300 MHz	AIX 5.1C
Intel Linux IA64	NEC AzusA 800MHz	Red Hat Linux 2.4.7-nec1.1p1
Intel Linux IA32	HP X4000 P4 Xeon 1700 MHz	MSC.Linux 2.4.6-1msc-smp
NEC	SX-6	SuperUX 12.1
SGI	Origin 3000 / 600 MHz	IRIX64 6.5
Sun	UltraSPARC 750 MHz	Solaris 8

• DMP is only recognized in SOL 101, 103, 108, and 111.
• Each task requires enough memory for the size of executable, the amount specified by the MEM parameter, and the amount needed for system I/O buffers. On true distributed systems, this amount is on each node. On SMP systems, all the tasks are on the same node.
• ACMS is very good for analyses with LOTs of modes (xx0cmd2), but is is very slow on analyses with few modes (xx0dmd0 with ACMS and DMP=2 took too long on most systems to even run and so results were not shown).
• In SOL 108 analyses, parallel efficiency increases as the number of frequencies (power of 2 is ideal) increase.

Hardware Details:

• Compaq

      Model:            Compaq AlphaServer GS160 (CC-NUMA)
      CPU:              16 X 1 GHz Alpha EV68
      OS Level:         UNIX 5.1 
      Memory:           64Gb
      Disk:             4 disk stripe set: Each disk stripe set had 10 18GB disks on
                        2 Ultra3 SCSI controller.
      

• Compaq

      Model:            Compaq AlphaServer SC (8 Nodes)
      CPU:              4 833MHz Alpha EV68 processors 
      OS Level:         UNIX 5.1 
      Memory:           4 nodes had 8Gb, 4 had 4 Gb
      Disk:             10 18.2GB disks on 2 Ultra3 SCSI controller striped using
                        LSM software.
      Interconnect Fabric: Quadrics
      

• Cray

      Model:            SV1ex (32 cpus)
      CPU:              SV1ex 500MHz
      OS Level:         UNICOS 10.0.0lf
      Memory:           2 Gbyte (4 GW) memory
      Disk:             96 Gbyte (12 GW) SSDI, configured with:
                        /ssd   4 Gbytes (0.5 GW)
                        swap  32 Gbytes (4.0 GW)
                        sds   60 GBytes (7.5 GW)
      

• HP

      Model:            Superdome 64-way 9000/800 (16 cells, 64 cpus)
      CPU:              PA-8700 750MHz
                        1.5MB data cache and 0.75MB I-cache
      OS Level:         HP-UX B.11.11
      Memory:           256 GB Real memory
                        238 GB Swap space
                        16 TB virtual address space
      Disk:             16 fc10s with 32 fiberchannel adapters and 159 disks
                        in single 2TB vxfs filesystem with dynamic buffer cache
                        set to 5% min, 60% max
      

• IBM

      Model:            p690 Turbo (32 cpus)
      CPU:              POWER 4 1300 MHz
      OS Level:         AIX 5.1C
      Memory:           96 GB Real memory
      Disk:             Sngle scratch JFS file system was striped over 96 (9GB)
                        SSA disks on 12 adapters.
      

• intel Linux IA64

      Model:            NEC Express5800/1160Xa (AzusA)
      CPU:              Itanium 800Mhz 
      OS Level:         Red Hat 2.4.7-nec1.1p1
      Memory:           16 Gb
      Bus:              33 MHz 64-bit PCI BUS
      Misc:             Processor Affinity control was used for more information regarding
                        Processor Affinity control please contact Cormac Garvey
                        (cgarvey@atcc.necsys.com).
      

• intel Linux IA32

      Model:            HP 4000
      CPU:              P4 Xeon (1700 MHz)
      OS Level:         MSC.Linux 2.4.6-1.msc-smp 
      Memory:           3 Gb
      Disk:             3X36Gb Software striped IBM Ultrastar DDYS-T36950
      

• NEC

      Model:            SX-6
      CPU:              SX-6 Vector processor (Max 8 GFLOPs per processor) Clock 2Ns
      OS Level:         SuperUX 12.1
      Memory:           64 GB (Real)
      Disk:             NEC Polestar Disk four-way stripping
      

• SGI

      Model:            Origin 3000 (16 cpus)
      CPU:              R1400A 600MHz
      OS Level:         IRIX64 6.5.14f
      Memory:           32 Gb
      Disk:             8 lun TP9400 RAID file system.
      

• SUN

      Model:            UltraSPARC 750 MHz
      OS Level:         Solaris 8