RE: [Fipa-os-developers] FIPA-OS failures under very heavy load

[Alan T. <ala...@em...>]

Hi Andres

->

> -----Original Message-----
> From: fip...@li...
> [mailto:fip...@li...]On Behalf Of
> Andres Ederra
> Sent: 26 August 2001 10:17
> To: fip...@li...
> Subject: [Fipa-os-developers] FIPA-OS failures under very heavy load
>
>
> Hello!
>
> Today i needed to make some test with several agents in a quite old
> computer(a k6 233 64MbRam)
> with several agents and i got problems.
>
> When i launched the whole platform (df,ams&acc) and 3 more agents(2 of
> them quite big and resource hungry).
> some messages we unable to be delivered.
>
>
> I got the error message that appears when you try to send a message to
> an agent that is not running.
> but i'm pretty sure that my agents are running ok, very slowly, but they
> are running.

Hmmm.... this doesn't sound good :(

> I did the test using a Win98 (with the scripts fixed to work with
> win98), and later with Linux, and i had the same problems.
> After i did the very same test on a "big machine" and i had no problems.
>
> It seems to me that there is some kind of  internal timeout in the MTS,
> that launches an error when an agent takes too long to answer (for
> example becouse it doesnt exist and it won't answer or when there exist
> a high system load).
> is that rigth? if it is, what class should i look to modify the timeout
> (im not familiar with the Fipa-OS MTS :( )

The MTS doesn't have any explicit timeouts defined - however their are
definately socket timeouts at the TCP layer of a TCP/IP protocol stack, and
I'm not sure about RMI (I'd guess it relies upon exceptions thrown from the
Socket classes to detect connection timeouts though).  Perhaps the VM is
becoming so overloaded that incoming socket connections aren't being
answered in time?  The precise reason should be contained in a DIAGNOSTICS
file if this is the case....

Out of interest, are all the agents running in one VM, or spread across
multiple VM's?

Thanks,
Alan

> If it's not a problem of the MTS classes it could me something of the
> rmi...
> Any clue?
>
> Thanx in advance
>
>     Andres Ederra
>
>
> _______________________________________________
> Fipa-os-developers mailing list
> Fip...@li...
> http://lists.sourceforge.net/lists/listinfo/fipa-os-developers
>
>