Wednesday, May 27, 2015

More Eclipse Tooling enhancements

The biggest complaint from our customers about the eclipse tooling for B*MS is that the cost of entry is too high; not only must a user be familiar with several different technologies, such as Git, maven, REST services and how these technologies are exposed by the eclipse tooling, but s/he must also understand the various Drools and jBPM configuration and definition files. Since there are only a few user-friendly/graphical editors that hide underlying file details, the user must become familiar with most of these file formats, and where in the Project or Repository hierarchy the file resides.

One of the enhancements I have been working on will hopefully ease some of this burden by providing a "navigator" similar to the Eclipse Project Explorer, but designed specifically for Drools/jBPM projects (see below).
At the root of this tree viewer are the app servers that have Drools/jBPM installed. Servers are managed (start, stop, debug) from the WST Servers view. At the next level is the Organizational Unit, then Repositories and finally Projects. Essentially, this viewer mimics the web console with the addition of multiple servers.

The tree structure is cached whenever a connection to the server can be established. This allows the view to be used in "offline" mode if the server is down or network connection is unavailable. When the server is available again, the viewer synchronizes its cache with the server.

Repositories are automatically cloned, and Projects are imported as they are requested by the user with a context menu action.

I'm still in the design/experimenting phase right now, so if there's a feature you'd like to see, or if you have suggestions for improving this interface please post your comments here.

You can also see a related post, showing my work on improving the wizards and runtime generation and configuration.

Share/Bookmark

Improved Drools & jBPM Eclipse wizard

Bob has been working on improving our Drools & jBPM Eclipse wizards.

  • The user no longer needs to create runtimes. They can now be created automatically on the fly by the new project wizard.
  • The project wizard will now list examples from the github repository and allow them to be selected and dowloaded as part of the wizard.
You cans see a video for this here:

Currently all the downloadable examples are jBPM, we still need to migrate the Drools examples over to this repository format.

Mark

Share/Bookmark

Wednesday, May 20, 2015

A Comparative Study of Correlation Engines for Security Event Management

This just paper came up on my google alerts, you can download the full text from ResearchGate.
"A Comparative Study of Correlation Engines for Security Event Management"

 It's an academic paper, published in the peer reviewed journal.
"10th International Conference on Cyber Warfare and Security (ICCWS-2015)"

Th paper is evaluating the correlation performance for large rule sets and large data sets in different open source engines. I was very pleased to see how well Drools scaled at the top end. I'll quote this from the conclusion and copy the results charts.
"As for the comparison study, it must be said that if the sole criteria was raw performance Drools would be considered the best correlation engine, for several reasons: its consistent behaviour and superior performance in the most demanding test cases."

In Table 2 (first image) we scale form 200 rules to 500 rules, with 1mil events with almost no speed loss - 67s vs 70s.

In Table 1 (second image) our throughput increases as the event sets become much larger.

I suspect the reason why our performance is less for for the lower rule and event set numbers, is due to the engine initialisation time for all the functionality we provide and for all the indexing we do. As the matching time becomes large enough, due to larger rule and data sets, this startup time becomes much less significant on the over all figure.





Share/Bookmark

Tuesday, May 12, 2015

Validation and Verification for Decision Tables

The decision tables are getting even more improvements than the UI work Michael has been working on.
Zooming and Panning between Multiple Huge Interconnected Decision Tables
Cell Merging, Collapsing and Sorting with Multiple Large Interconnected Decision Tables

I am currently working on improving the validation and verification of the decision tables. Making it real time and improving the existing V&V checks.

Validation and verification are used to determine if the given rules are complete and to look for any bugs in the dtable authors logic. More about this subject.

Features coming in the next release


Real time Verification & Validation

Previously the user had to press a button to know if the dtable was valid or not. Now the editor does the check in real time, removing the need to constantly hit the Validate-button. This also makes the V&V faster, since there is no need to validate the entire table, just check how the change of a field affected the rest of the table.




Finding Redundancy 

To put it simple: two rows that are equal are redundant, but redundancy can be more complicated. The longer explanation is: redundancy exists when two rows do the same actions when they are given the same set of facts.

Redundancy might not be a problem if the redundant rules are setting a value on an existing fact, this just sets the value twice. Problems occur when the two rules increase a counter or add more facts into the working memory. In both cases the other row is not needed.




 

 

Finding Subsumption

Subsumption exists when one row does the same thing as another, with a sub set of the values/facts of another rule. In the simple example below I have a case where a fact that has the max deposit below 2000 fires both rows.

The problems with subsumption are similar to the case with redundancy.






Finding Conflicts

Conflicts can exists either on a single row or between rows.
A single row conflict prevent the row actions from ever being executed.

Single row conflict - second row checks that amount is greater than 10000 and below 1






Conflict between two rows exists when the conditions of two rules are met with a same set of facts, but the actions set existing fact fields to  different values. The conditions might be redundant or just subsumptant.

The problem here is, how do we know what action is made last? In the example below: Will the rate be set to 2 or 4 in the end? Without going into the details, the end result may be different on each run and with each software version. 
Two conflicting rows - both rows change the same fact to a different value

 

Reporting Missing Columns

In some cases, usually by accident, the user can delete all the condition or action columns.

When the conditions are removed all the actions are executed and when the actions columns are missing the rows do nothing.
The action columns are missing
The condition columns are missing










What to expect in the future releases?


Better reporting

As seen on the examples above. Reporting the issues is currently poor.
The report should let the user know how serious the issue is, why it is happening and how to fix it.

The different issue levels will be:
  • Error - Serious fault. It is clear that the author is doing something wrong. Conflicts are a good example of errors.
  • Warning - These are most likely serious faults. They do not prevent the dtable from working, but need to be double checked by the dtable author. Redundant/subsumptant rules for example, maybe the actions need to happen twice in some cases.
  • Info - The author might not want to have any conditions in the dtable. If the conditions are missing each action gets executed. This can be used to insert a set of facts into the working memory. Still it is good to inform that the conditions might have been deleted by accident.  

 

Finding Deficiency

Deficiency gives the same kind of trouble that conflicts did. The conditions are too loose and the actions conflict.

For example:
If the loan amount is less than 2000 we do not accept it.
If the person has a job we approve the loan.
The problem is, we might have people with jobs asking for loans that are under 2000. Sometimes they get them, sometimes they do not.


 

Finding Missing Ranges and Rows

Is the table complete? In our previous examples we used the dtable to see if the loan application gets approved. One row in the dtable should always activate, no matter how the user fills out his loan application. Either rejecting or approving the loan or else the applicant does not get a loan decision.
The goal of the V&V tool will be to find these gaps for the dtable author.

 

Finding Cycles

The actions can insert new facts and the conditions trigger the actions when new facts are inserted. This can cause an infinite number of activations.
This issue is a common mistake that the goal is to pick it up in the authoring phase with the V&V tool.


Share/Bookmark