Subject: RE: [jade-develop] BlueJADE package available
From: Robert Kessler (kessler@cs.utah.edu)
Date: Tue Sep 17 2002 - 19:29:45 MET DST
Lots and lots of issues. I'll pipe in with a few:
1) Using JADE in an AS is extremely powerful. What it allows you
to do is add a level of robustness to an MAS that you have developed.
This comes from the ability to automatically restart agents that have
died for some unexpected reason. If as a community we move into the
world where agents are up 24/7 doing stuff for us, we really need a
robust environment around the agent system to ensure that 24/7 is a
reality. My experience running under the previous BlueJade system that
was built on top of HPAS was that this really worked. There were times
that an agent would crash (for whatever reason) and the system would
restart it. But the problem is being able to recognize a crash.
Sometimes we have seen behaviour where the agent simply quits
responding. That implies that we need some kind of watchdog facility
that allows an external force to decide that the agent is no longer
viable and must be restarted. Of course, then you need a watchdog of
the watchdog. So, robustness seems to me to be really important if we
are to move MAS into doing full time useful work. [Don't forget, what
an AS is designed to do is keep the web services that it is providing up
- so this is a nice marriage.]
2) As you know, we have been playing in the world of web services
and agent system interactions. Some say that agents are unnecessary,
you can do it all with web services. Others say that web services are
not powerful enough and it should all be done with agents. There are
legitimate arguments on both sides (this is independent of talking about
the AOSE issues which I think is another interesting domain of study for
agents, but that really is independent of this issue). However, my
feeling is that we need both and we need them to be "merged". There are
capabilities that agents have that web services do not currently enjoy
(autonomy, proactiveness, etc.). As web services grow and become more
capable it is natural to see that they are going to need more and more
agent capabilities. Could we use a more high-level declarative model
for exchanging information between web services? Yes. Could we use the
ability for a web service to be proactive? Yes. . But we certainly
aren't any where near this kind of level of integration. So, from my
view of the world, using an AS which is designed to handle web services
and that is enhanced to also handle agents is a perfect place to
research this important question of web services and agents.
Bob.
-----Original Message-----
From: jade-develop-admin@sharon.cselt.it
[mailto:jade-develop-admin@sharon.cselt.it] On Behalf Of Bellifemine
Fabio
Sent: Tuesday, September 17, 2002 11:05 AM
To: jade-develop
Subject: RE: [jade-develop] BlueJADE package available
Dear JADE users and agent programmers,
as a follow up of the announcement of Dick Cowan, on behalf of the JADE
team, I would like to open a new thread of discussion.
The usage of the jade.wrapper API allowed Dick to integrate JADE as a
service of an Application Server. This is surely a valid step towards
the integration with the web world but the question that we have is the
following: what is the final goal that we should try to reach, i.e. what
should be done after this step?
In order to get answers to this difficult question for the entire agent
community, of course we should try to start from the analysis of the
requirements.
1. What are the advantages that a developer of multi-agent systems, in
particular a JADE programmer, gets in using JADE integrated into an
Application Server (e.g. in JBoss)? Why he should decide to do that?
Maybe simply because he needs to deploy its application on a Web server.
And then, how we would like to partition the tasks between the
Application Server and the JADE environment?
2. The other way around. What are the advantages that a developer of Web
applications gets in using JADE?
For instance, the fact that he can implement a control logic with a life
cycle independent of a Web session and where independent threads can
easily communicate each other. Example: a user logs into a web site,
fills a form with indications about an item to sell/buy, and then he
logs off; the web session terminates but a JADE agent is created with
the goal of accomplishing the user task.
I would really appreciate feedback and input from the entire community.
Kind regards, Fabio.
Lots and lots of issues. I’ll pipe in with a few:
1) Using JADE in an AS is extremely powerful. What it allows you to do is add a level of robustness to an MAS that you have developed. This comes from the ability to automatically restart agents that have died for some unexpected reason. If as a community we move into the world where agents are up 24/7 doing stuff for us, we really need a robust environment around the agent system to ensure that 24/7 is a reality. My experience running under the previous BlueJade system that was built on top of HPAS was that this really worked. There were times that an agent would crash (for whatever reason) and the system would restart it. But the problem is being able to recognize a crash. Sometimes we have seen behaviour where the agent simply quits responding. That implies that we need some kind of watchdog facility that allows an external force to decide that the agent is no longer viable and must be restarted. Of course, then you need a watchdog of the watchdog… So, robustness seems to me to be really important if we are to move MAS into doing full time useful work. [Don’t forget, what an AS is designed to do is keep the web services that it is providing up – so this is a nice marriage.]
2) As you know, we have been playing in the world of web services and agent system interactions. Some say that agents are unnecessary, you can do it all with web services. Others say that web services are not powerful enough and it should all be done with agents. There are legitimate arguments on both sides (this is independent of talking about the AOSE issues which I think is another interesting domain of study for agents, but that really is independent of this issue). However, my feeling is that we need both and we need them to be “merged”. There are capabilities that agents have that web services do not currently enjoy (autonomy, proactiveness, etc.). As web services grow and become more capable it is natural to see that they are going to need more and more agent capabilities. Could we use a more high-level declarative model for exchanging information between web services? Yes. Could we use the ability for a web service to be proactive? Yes. … But we certainly aren’t any where near this kind of level of integration. So, from my view of the world, using an AS which is designed to handle web services and that is enhanced to also handle agents is a perfect place to research this important question of web services and agents.
Bob.
-----Original Message-----
From:
jade-develop-admin@sharon.cselt.it [mailto:jade-develop-admin@sharon.cselt.it] On Behalf Of Bellifemine Fabio
Sent: Tuesday, September 17, 2002
11:05 AM
To: jade-develop
Subject: RE: [jade-develop]
BlueJADE package available
Dear JADE users and agent programmers,
as a follow up of the announcement of Dick Cowan, on behalf of the JADE team, I would like to open a new thread of discussion.
The usage of the jade.wrapper API allowed Dick to integrate JADE as a service of an Application Server. This is surely a valid step towards the integration with the web world but the question that we have is the following: what is the final goal that we should try to reach, i.e. what should be done after this step?
In order to get answers to this difficult question for the entire agent community, of course we should try to start from the analysis of the requirements.
1. What are the advantages that a developer of multi-agent systems, in particular a JADE programmer, gets in using JADE integrated into an Application Server (e.g. in JBoss)? Why he should decide to do that?
Maybe simply because he needs to deploy its application on a Web server. And then, how we would like to partition the tasks between the Application Server and the JADE environment?
2. The other way around. What are the advantages that a developer of Web applications gets in using JADE?
For instance, the fact that he can implement a control logic with a life cycle independent of a Web session and where independent threads can easily communicate each other. Example: a user logs into a web site, fills a form with indications about an item to sell/buy, and then he logs off; the web session terminates but a JADE agent is created with the goal of accomplishing the user task.
I would really appreciate feedback and input from the entire community.
Kind regards, Fabio.
This archive was generated by hypermail 2a22 : Tue Sep 17 2002 - 19:28:18 MET DST