Discussion:
[jade-develop] Jade undergoing a lag in message delivery (use-case explained in body)
krishna rangaiah
2014-02-18 07:38:50 UTC
Permalink
Hi Guys,

We have been experiencing some bottlenecks in the dispatch of messages
between agents in JADE. The context of the problem is as follows;

*Test Setup*
Given a pair of communicating agents (Agent A, Agent B), Agent A sends an
ACL message to Agent B which records the timestamp when it receives the
message. This is done to estimate the impact that variable ACL message size
in the range of {100KB, 200KB, .....5MB} has on the platform.

*Observation*
We have observed that when the ACL message size is <= 100KB, the message
manager is able to hold up just fine i.e. Even when the number of agents is
increased such that there are 'k' pairs of communicating agents in the
system that exchange messages <=100KB between them, the platform functions
normally.

However, when the message size is increased the platform begins to slow
down and at a size of >=300KB, the lag becomes too heavy and impacts all
the other agents in the system. i.e. If even one pair of communicating
agents exchange messages >=300KB, it is observed that remainder agents in
the system function with huge lags.

Some questions we have in this regard are:-

1. Is there any load balancing mechanism available at the message manager's
disposal to handle a scenario where message sizes increase?
2. Can we alter the specifications of the message manager and dispatch
mechanism while not having to override the MTP and use our own message
transfer protocol?
3. This behavior is a direct precursor to the queuing of an agent's inbox
and we believe that addressing this will also address the queuing problem
to a certain extent. We'd like to know your thoughts on this?

Thanks in advance for all your help,

Regards,
Krishna K
--
I am Only One, but still I am one; I cannot do everything, but still I can
do something; and because I cannot do everything, I will not Refuge to do
something that I CAN DO.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://avalon.cselt.it/pipermail/jade-develop/attachments/20140218/4b854396/attachment.html>
Caire Giovanni
2014-02-19 10:52:55 UTC
Permalink
Hi,

Which version of JADE are you using? Up to version 4.3 there was an "odd protection mechanism" that caused the whole system to become very slow when a lot of big messages where exchanged. Such mechanism was fixed in 4.3.1 released the 6th of December 2013.
So, if you performed your tests with JADE 4.3 (or an earlier version), this can be the reason for your observation. In that case can you please run your tests again with 4.3.1 and let us know?

Bye,

Giovanni

Da: jade-develop-bounces at avalon.tilab.com [mailto:jade-develop-bounces at avalon.tilab.com] Per conto di krishna rangaiah
Inviato: marted? 18 febbraio 2014 08:39
A: jade-develop at avalon.tilab.com
Oggetto: [jade-develop] Jade undergoing a lag in message delivery (use-case explained in body)

Hi Guys,

We have been experiencing some bottlenecks in the dispatch of messages between agents in JADE. The context of the problem is as follows;

Test Setup
Given a pair of communicating agents (Agent A, Agent B), Agent A sends an ACL message to Agent B which records the timestamp when it receives the message. This is done to estimate the impact that variable ACL message size in the range of {100KB, 200KB, .....5MB} has on the platform.

Observation
We have observed that when the ACL message size is <= 100KB, the message manager is able to hold up just fine i.e. Even when the number of agents is increased such that there are 'k' pairs of communicating agents in the system that exchange messages <=100KB between them, the platform functions normally.

However, when the message size is increased the platform begins to slow down and at a size of >=300KB, the lag becomes too heavy and impacts all the other agents in the system. i.e. If even one pair of communicating agents exchange messages >=300KB, it is observed that remainder agents in the system function with huge lags.

Some questions we have in this regard are:-

1. Is there any load balancing mechanism available at the message manager's disposal to handle a scenario where message sizes increase?
2. Can we alter the specifications of the message manager and dispatch mechanism while not having to override the MTP and use our own message transfer protocol?
3. This behavior is a direct precursor to the queuing of an agent's inbox and we believe that addressing this will also address the queuing problem to a certain extent. We'd like to know your thoughts on this?

Thanks in advance for all your help,

Regards,
Krishna K


--
I am Only One, but still I am one; I cannot do everything, but still I can do something; and because I cannot do everything, I will not Refuge to do something that I CAN DO.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://avalon.cselt.it/pipermail/jade-develop/attachments/20140219/7d815844/attachment.html>
Loading...