From: Eric Hodel <drbrain@...7.net>
Date: 2006-03-29T04:12:38+09:00
Subject: Re: Tracking DRbObject References?

On Mar 27, 2006, at 5:27 PM, Daniel Nugent wrote:

> Does anyone know of any work done on keeping track of the number of
> the existing DRbObject references to a local object?

IdConv would be the thing that does this.

> It seems that if I can keep track of the propogation of DRbObjects
> (adding calls to a service on the server where the local server that
> update the distributed reference count), that it should be sort of
> trivial to release the local object only when all distributed
> references are gone.
>
> On #ruby-lang it was suggested that this might be premature
> optimization, however, I am sure that this will be necessary.

!!!BIG, RED WARNING FLAG!!!

Program based on necessity, not on inferred necessity!

> The purpose of the system I am writing is to provide a framework  
> for Actor
> computation.  One of the programming techniques in the Actor model is
> to produce a large number of worker actors on a machine with many
> processors for the express purpose of computing a small part of a much
> larger computation.  However, another technique is to create a smaller
> number of actors that you will keep around and possibly propogate
> those references to other actors in the system.  So you can see how
> the distributed system could easily become littered with
> undiscoverable objects that would sit around wasting resources if they
> aren't garbage collected, yet we can't necessarily toss them away
> right after we use them.
>
> If there's anything that you think might be helpful, I'd be very  
> appreciative.

There are two better options for tracking distributed objects.

Switching to a different IdConv like NamedIdConv.

Using a TupleSpace to manage the distributed objects in your system.

-- 
Eric Hodel - drbrain@segment7.net - http://blog.segment7.net
This implementation is HODEL-HASH-9600 compliant

http://trackmap.robotcoop.com