Manners Performance Benchmarks Results

The Manners problem involves assigning each guest a seat in a single row such that they sit next to guests of the opposite sex with whom they share a hobby.

The results show that Eclipse solves the problem for 64 guests in one twentieth (1/20) of a second.

The performance shown uses interpreted rules that can be changed dynamically.

No C++ or Java compilation is necessary.

This means that dynamic rule changes can be loaded without rebooting a server, for example.

This compares to published results in which ILOG requires over 200 seconds on an Ultra Sparc.

Download the Manners Benchmark video
This video is a Lotus ScreenCam executable that includes the player - the video will run on any Windows PC without additional software.

Additional Manners Benchmark information is available in an InfoWorld article and at The University of Texas.

Java rules engine benchmarks

JESS, the Java Expert System Shell, is available as a free download (including its complete Java source code) from the Department of Energy. JESS is derived from CLIPS (NASA's C Language Integrated Production System). JESS and CLIPS use the syntax and algorithms that were first introduced in 1984 by Paul Haley and later enhanced within the Eclipse product line. Despite being implemented in Java, JESS typically runs faster than CLIPS which is implemented in C, particularly as the amount of data involved increases.

100% Java and Eclipse

Eclipse engines execute within 100% Java applications as native code. Although native implementation helps Eclipse run faster in Java, Eclipse is more than a few times faster since it uses the most functionally advanced derivative of the Rete Algorithm available.

Relative Performance

The following table shows how one of the fastest 100% Pure Java rules engine stacks up against Eclipse on a suite of publicly available benchmarks:

Performance Benchmarks Results

Eclipse Inference Engines:

  • integrate seamlessly within Java applications
  • are typically an order of magnitude more efficient than other engines
  • offer performance that scales to thousands of rules without degradation
  • can serve thousands of clients with effectively no latency