This document summarizes experiments comparing computer clusters built with different hardware for mechanical design workloads. The experiments found that a cluster of Dell OptiPlex PCs running Windows NT provided comparable stability and multitasking performance to a more expensive Sun workstation, while scaling better on parallel applications. NT also showed better memory management. Overall, the Dell cluster provided the best performance and value for mechanical design environments.

1. Building the Best Computer
Clusters for Mechanical Design
Environments
Daniel Chan and Ed Huott
e
GE Corporate R&D Center

2. Key Challenges
• A Highly Diverse Environment
– Many Different Businesses
– Engineering, Chemistry, Financial, Services
• Short-Term Technical Support
– Quick Turnaround
• Costs
– Downtime, System Admin., Software
Licenses, Productivity
e

3. Design Requirements
• Web-Centric
– Simplify Job Submission in a Heterogeneous
Environment
– Access from Anywhere and Anytime
• Minimize Complexity and Cost
• Stable and Highly-Available
– Isolate Network, NFS and Insufficient Disk
Space Problems, Software License
Management e

4. Reasons for Having a
Heterogeneous Environment
• Performance
• Costs
• Legacy Applications
e

5. Let’s Do An Experiment
NT 4.0 Unix/Solaris 2.6
GE Corporate Standard Sun Ultra 10
PIII 450 MHz UltraSPARC 440 MHz
256 MB RAM 256 MB RAM
$1400 $5000
Use Perl script CFX 5 Extract wall
to launch 1, 2 and 23,862 Nodes clock times from
3 concurrent jobs, 102 MB Required output files then
respectively, for use Minitab to
30 times analyze data
Test stability, multitasking and paging capabilities
e

6. Results For Sun Ultra 10
1 job
Histogram of C4, with Normal Curve
2 jobs
Histogram of C4, with Normal Curve
8 30
7
6
20
Frequency
5
Frequency
4
3
10
2
1
0 0
1700 1710 1720 1730 1740 1750 1760 3320 3340 3360 3380 3400 3420 3440 3460 3480 3500
C4 C4
Histogram of C4, with Normal Curve
50
40
Frequency
3 jobs
30
e
20
10
0
4000 5000 6000
C4

7. Results for Dell OptiPlex G1x
Histogram of TOTAL_CLOCK_TIME, with Normal Curve
1 job 2 jobs
Histogram of C4, with Normal Curve
20 40
30
Frequency
Frequency
10 20
10
0
0
1924 1925 1926 1927 1928
3900 3950 4000
TOTAL_CLOCK_TIME
C4
3 jobs
Histogram of TOTAL_CLOCK_TIME, with Normal Curve
50
40
e
Frequency
30
20
10
0
5800 5850 5900

8. Scorecard for Sun Ultra 10
USL − µ 01µ
.
Z= =
σ σ
Mean Standard Z
Deviation
1 job 1724 15.9 10.8
2 jobs 3442 26.1 13.2
(2X)
3 jobs 5144 350.2 1.5
(3X)
Paging
e

9. Scorecard For Dell Optiplex GX1
USL − µ 01µ
.
Z= =
σ σ
Mean Standard Z
Deviation
1 job 1926 1.2 161
2 jobs 3909 15.6 25
(2.1X)
3 jobs 5828 26.4 22
(3X)
e

10. Summary of Results
• The SUN workstation is about 10% faster, but
nearly 5 times as expensive
• Both systems are just as stable for a period of
one week
• NT can multitask
• NT can page and appears to have “better”
memory management scheme e

11. APNASA Parallel Performance
308,115 Grid Points
1400
PII 333 MHz Cluster
1200
1000
TIME (SEC)
800
600
400 SGI Origin 2000
R10K 195 MHz
200 PIII 550 MHz Xeon
e
2 MB Cache, 8-Way SMP
-2 -1 0 1 2 3 4 5 6 7 8 9 10 11
No. of Processors

12. APNASA Parallel Performance
308,115 Grid Points
IDEAL
9
8
7 PII 333 MHz Cluster
6 SGI Origin 2000
R10K 195 MHz
5
Speed Up
4
3 PIII 550 MHz Xeon
2 2 MB Cache, 8-Way SMP
1
e
0
0 1 2 3 4 5 6 7 8 9 10 11
No. of Processors

13. CFX PERFORMANCE ON A
VARIETY OF COMPUTER
PLATFORMS
556323 VERTICES
SGI SMP
81118 VERTICES
CPU TIME (SECS)
4 PII 333 MHz NT
10
81118 VERTICES
WORKSTATIONS
81118 VERTICES
SGI SMP
3
10
23862 VERTICES 23862 VERTICES
WORKSTATIONS SGI SMP
0.8 1
0.91 2 3 4 5 6 7 8 910
NUMBER OF PROCESSORS
10 20
e

14. ANSYS Performance on a Variety of
Computers
Sun Enterprise 450
Ultra II 400 MHz ELAPSED TIME
HP J5000, PA8500 400 MHz CPU TIME
Compaq SP700, PIII 500 MHz
SGI 320, PIII 500 MHz
Dell Optiplex GX1, PII 450 MHz
Dell 6300, PIII 500 MHz Xeon
Compaq DS20, 500 MHz EV6
SGI O2K R10K 250 MHz
e
0 1 2 3 4 5 6 7 8 9 10 11 12
CPU TIME (HRS)
160,000 ELEMENTS WITH 2D CONTACT
2.1 GB OF OUTPUT AND 615 MB PAGE FILE

15. eComputing Architecture
Web-Enabled Environment
Load Sharing Facility
Quality Tools
SGI SC
Remote Access from
HP Cluster
Anywhere Anytime
(Globalization)
NT Cluster
Field and Customer Data DM Cluster
(Services and eCommerce)
user-transparent work
load management
e

16. Key LSF Challenges
• Implicit Assumptions of A Homogeneous
Environment
– Uniform View of NFS
– Uniform Mounting of User Home Directory
– Unix Bias
• Wholesale Migration of User
Environment Variables from Submit to
Execution Hosts
e

17. A Web-Centric Job Management
System
https Web NFS Job Spec.
Web Clients Files
Server
LSF Proxy
NFS or SMB
ftp
Execution Hosts
Job Directories LSF Cluster
e

18. Screen Shot of Web Interface
Job Source Files
Job Output Destination
Command to Run on
Execution Host
e
Files Sent to
Output
Destination

19. Process, Data Flow and Security
for Web-Based Jobs
https Apache
Web Client Job Spec
(invoke_bsub.pl)
Authenticated access to
setuid script File system
permissions
bsub
ftp Exec Host
Job Dir (lsf_exec.pl)
LSF Cluster
e

20. Normalized Job Execution Wrapper/
Environment
• Standard job execution wrapper that runs on all execution hosts.
• Key "ingredients": Perl (with Net::FTP package), wget, subset of
"standard" Unix utilities (e.g. /bin/sh, cp, mv, etc.)
– (Note: Makes use of Cygwin tools on Windows NT. See http://
sources.redhat.com/cygwin/)
• Reads Job Specification File to determine:
– Source of remote job directory
– Command to run o Destination for job output file(s)
• Pre-determined top level (local) directory for jobs on each execution
host.
• Remote jobs are migrated by FTP (wget) to unique sub-directories
under top level.
e
• Specified command is run in migrated job directory.
• Output files are transferred by FTP to specified destination.

21. Summary
• Leveraging Low-Cost NT Solution
• Shield Users from Complexity
– productivity gain
– better centralized resource management
• Enterprise Resource Planning
– compute cycles
– software licenses
e