Enterprise Java

Drools 6 Performance with the PHREAK Algorithm

Drools 6 introduces a new lazy matching algorithm. The details of that algorithm have been covered in two previous blogs:

The first article discussed performance and why the batch and lazy aspects of the algorithm, make it hard to compare.
 
 
 

“One final point on performance. One single rule in general will not evaluate any faster with PHREAK than it does with RETE. For a given rule and same data set, which using a root context object to enable and disable matching, both attempt the same amount of matches and produce the same number of rule instances, and take roughly the same time. Except for the use case with subnetworks and accumulates.

PHREAK can however be considered more forgiving that RETE for poorly written rule bases and with a more graceful degradation of performance as the number of rules and complexity increases.

RETE will also churn away producing partial machines for rules that do not have data in all the joins; where as PHREAK will avoid this.

So it’s not that PHREAK is faster than RETE, it just won’t slow down as much as your system grows!”

More recently the OptaPlanner team started benchmarking the same set of rules between ReteOO and Phreak: Which rule engine algorithm is faster: ReteOO or Phreak? 

They found that that three tests were 20% faster, and one test 4% slower. A user left a comment, for that article, relating at 17% performance difference.

phreakperf

The OptaPlanner team have invested time into making sure the way they write the rules do not hit any Rete walls.  They have removed many of the issues, such as multiple accumulates in a single rule.

One user was interested in what happens if you implement rules in a way known to cause problems with ReteOO, would it handle it more gracefully. They created 4 rules, each with two accumulates. The DRL file can be found here, and a copy of one of the rules is shown below:

rule gold_account
when
  account: Account()
  Number(this >= 50000) from accumulate(t: Transaction(source == account); sum(t.amount))
  Number(this >= 50000) from accumulate(t: Transaction(target == account); sum(t.amount))
then
  //System.out.println("Gold account: " + account);
end

The results were promising, a 400% (5x) performance gain for Phreak :) This is mostly enabled by the way that Phreak evaluates rules in batches, avoiding a lot of wasted work. It certainly shows we’ve achieved one of the aims quoted from the earlier paragraph:

“PHREAK can however be considered more forgiving that RETE for poorly written rule bases and with a more graceful degradation of performance as the number of rules and complexity increases.”

The algorithm so far has been designed for correctness, especially with regards to thread safety and future multi-core exploitation. So this is just the start, we still have plenty more optimisations and improvements to do.
 

Reference: Drools 6 Performance with the PHREAK Algorithm from our JCG partner Geoffrey De Smet at the Drools & jBPM blog.

Geoffrey De Smet

Geoffrey De Smet (Red Hat) is the lead and founder of OptaPlanner. Before joining Red Hat in 2010, he was formerly employed as a Java consultant, an A.I. researcher and an enterprise application project lead. He has contributed to many open source projects (such as drools, jbpm, pressgang, spring-richclient, several maven plugins, weld, arquillian, ...). Since he started OptaPlanner in 2006, he’s been passionately addicted to planning optimization.
Subscribe
Notify of
guest

This site uses Akismet to reduce spam. Learn how your comment data is processed.

0 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Back to top button