SCL Cluster Cookbook
HINT Benchmarks of Commodity Systems

·   Running HINT   ·
·   Cluster Cookbook Home   ·   Scalable Computing Lab   ·

HINT is a benchmarking program that provides a graph of system performance, not in bogus terms of MIPS or MFLOPS, but in a rigorously defined mathematical definition of work called QUIPS.

HINT determines the speed for a given machine specification and problem size. Computers typically are fastest on problems of small size and as the problem size grows, they slow down as they run out of fast memory and start using the main memory. They slow down even more if they have to access the disk. This performance curve is clearly visible from HINT generated data, and the curves are easily compared for different machines.

The graph below shows the HINT performance for a representative 200MHz Pentium Pro and 266MHz Pentium II system compared to a 500MHz Digital Alpha and an SGI R10000-based system.

Figure 1. HINT Comparison of Pentium Pro and Pentium II to Competive CPUs

Since we are concerned with clusters, it is interesting to compare clusters to supercomputers with shared memory or special processor interconnects. The graph below shows the HINT performance observed on a four-node Pentium Pro cluster connected through a Fast Ethernet switch compared to four nodes of an IBM SP-2 and four nodes of an SGI PowerOnyx. While HINT does not test the performance of the interconnect, it does show that the aggregate processor performance of the Pentium Pro rivals supercomputer-class machines.

Figure 2. Comparison of Multiple-Processor Systems

See the Running Hint page to see how to benchmark your systems with HINT.


Questions or comments? Send mail to ghelmer@scl.ameslab.gov
Copyright © 1997, 1998 All Rights Reserved by Scalable Computing Laboratory.
Disclaimer Notice
Maintained by ghelmer / Last updated on 06/10/98