that's one of the performance limitations with full fact replication
approach for distributed pattern matching. In a situation where a
significant percent of the facts always change, the most optimal method is
to replicate just the indexes. Most modern RETE implementation like JESS
already index the facts, so index replication allows the system to perform
index joins without the fact. It's similar to how databases do index joins
for partitioned tables.

I should state that I filed a patent for distributed RETE using index
replication back in 2004. As far as I know, that should be the most
efficient method of implementing distributed RETE and distributed reasoning.

peter

On Feb 20, 2008 9:12 AM, Mohd. Noor <[EMAIL PROTECTED]> wrote:

>
> Hi all,
> In the DJESS - how often the working memory/fact/repository has to be
> replicated?
> How about if my the information that I am going to fire is always change-
> so this is means that my distributed information are also need to change.
>
> Rgds
> mnoor
>
>
>

Reply via email to