Re: [Swarm-Support] Re: [Swarm-Modelling] xmalloc problem

[Marcus G. Daniels]

Steve Railsback wrote:

> Right, but: We (and apparently some others) find this xmalloc problem 
> with a variety of codes that do not leak memory a lot and work fine 
> until we either (a) simulate a lot more agents, or (b) run the program 
> for a really long time. 

..just a matter of what is "a lot" and whether a person really knows 
what that amount is that is leaked, if any.  Sounds like in your case 
you do know.  384M (the default limit) is certainly a limit one can hit, 
but the advantages of not hitting it by avoiding needless waste may be 
worth the trouble and have performance benefits on other platforms as 
well.  

In the case of Java (Jose's case), it is all too easy to waste memory by 
invoking Selectors on Java methods.  This causes Java object arguments 
to become backed by Objective C proxies and result in worsening wasted 
memory.  What should be used are FActions instead of Selectors.  These 
proxy can be observed accumulating if one takes a look.

So "make it bigger" can be necessary in some cases, but it is no panacea.