Re: [moca] Mobile Agents: Ten reasons for failure

[Massimo Merro <merro@xxxxxxxxxxxx>]

Dear Pablo,

thank you very much for your post!

> My knowledge about Mobile agents is very limited, however I believe 
> the point of view Mobile Agent presented here is different than the 
> point of view of calculi a' la Ambient.

> Ambients implicitly assume a common infrastructure that supports 
> ambients and their operations. 

well, Ambient is a process algebra to model Mobile Agents. So, of course 
many aspects are abstracted way. I do not understand exactly what you 
mean with a "common infrastructure". On the other hand, I hope we agree 
that the Mobile Agents we would like to model  are "those" indicated in 
Vigna's paper: "Mobile Components that can explicitly relocate 
themselves across the Network, usually preserving their exuction state 
across migration, a' la Telescript"

> > 5) Mobile Agents are difficult to authenticate and control
>
>
> Well, there is Proof Carrying Code to certify the code. Regarding 
> authentication, it guess depends on the application domain. A 
> authentication mechanism should be provided but establishing the 
> policy is a different issue and depends on each specific domain.

Yes,  PCC is a very useful technique. However, her the point is another, 
as Vigna says, apriori, an agent may
be associated with many identities:
a) The agent developer
b) the agents' code signer
c) the agent code dispatcher
d) the host the agent visited last...

which one should be used in which case?

> > 6) Mobile Agent can be brainwashed
>
>
> Well, yes. If a mobile agent is considered information, it can be 
> altered or simple erased. A cooperative infrastructure is always 
> needed, unless the mobile agent has hardware on its own. 

Yes, but this a DRAMATIC ISSUE! When I heard the first seminar on Mobile 
Agents, a typical example was that of some  travel agent visiting 
different sites looking for the best prices of some product and 
reporting home the result of its research.  Well, it seems to me that 
this kind of applications should never be implemented with mobile 
agents! It is definitely not safe for the travel agent to migrate to a 
new market place "with its best price in the pocket" asking for the best 
price proposed by the host. The host could simply "put its hands in the 
agent's pocket", changing the agent's best price, and then making its 
own price.

So, what kind of applications on mobile code would really benefit from 
adopting the Mobile Agent paradigm?

After an exchange of email with Lucian Wischik we agreed that we know 
only three real cases of mobile code:

(1) download of code in java and activex;
(2) upload of SQL programs;
(3) upload of Postscript programs.

However, none of them needs Mobile Agents. Does any of you know other 
examples?


Thanks again!

Regards,

--Massimo Merro.


 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The "models for mobility" mailing list     mailto:moca@xxxxxxxxxxxxxxx
http://www-sop.inria.fr/mimosa/personnel/Davide.Sangiorgi/moca.html