Europort Benchmarking

Europort created a benchmarking process designed to compare code scalability across different parallel architectures and for different problem sizes. Every application was benchmarked on at least two reference platforms with three different data sets. The testers ran these codes on each platform using progressively increasing numbers of nodes so as to assess the scalability of the code and estimate its parallel efficiency.

The benchmark graphs are usually flattening curves, showing that parallel efficiency generally falls off as you add more processors, until you reach a point where the extra speedup is not worth the cost of adding more CPUs. That's th e sort of economic calculation that benchmarking makes easier.

The Europort hardware reference platforms are:

• IBM SP2 (128 and 28 nodes)
• Meiko CS-2 (32 nodes)
• Parsytec GC PowerPlus (64 nodes)
• Silicon Graphics PowerChallenge Array (64 nodes)
• DEC Alpha "farm" (eight nodes)

---

Testing Across Applications

screen_link (62 Kbytes)

Test measures parallel efficiency of MaxHom application on up to 28 CPUs.