Re: [Chicken-users] To whom it may concern: Working on TinT

[F. Wittenberger]

Hello Kon,

how far are you come already?  I'm very interested!

What you've given in your mail is very, very close to the internals of
www.askemos.org .  To be precise that is (since Askemos means just the
API) the rscheme based implementation called BALL.

There's one major difference: Askemos processes are persistant and
replicated over a network in byzantine agreement, thus they continue to
work correctly while a minority of the nodes goes down or deliberately
cheat.

Askemos/BALL still depends on rscheme (I once did some work to port it
to chicken, but this is not yet finished).  I really want to move it to
chicken for several reasons.

BTW: I haven't been able to retrieve the Termite PDF (from LtU link) and
did not find any home page in google's mess.  Any good links to Termite?
Thanks

best regards

/jörg

Am Dienstag, den 14.02.2006, 12:06 -0800 schrieb Kon Lovett:
> Hello List,
> 
> I am working on TinT (Tint is not Termite), a Chicken specific  
> Termite workalike. Alike in the sense that is generally shares syntax  
> & semantics, but I feel under no constraint to be 100% faithful. (How  
> could I be, Termite is still evolving.)
> 
> Concepts
> 
> - Follows "impure" Actor model.

I don't understand the "impure" here.  Otherwise: same.

> - Messages are sent & received. A message is any serializable scheme  
> object. Message objects are immutable.

dito

> Non-serializable objects are proxied by a process.

NYI (not yet implemented)

> All messages are typed with a parallel message typing system. (Non- 
> Termite)

?

> - Message Receiver is a process. Message Sender is a process.
>
> - A process belongs to a Process Runner.  A process runner is on a  
> node. (Non-Termite)
> 
> Node corresponds to Machine. Process Runner to OS Process.
>
> - A node is local or remote. All local nodes share physical  
> resources, especially memory. A remote node only shares a network.
> 
> The current process runner on a local node is called the This Node.

All the same.

> - A this node process is an active procedure (SRFI-18 threads).

Different implementation right now.  We need to improve the byzantine
agreement protocol, which will probably be better implemented as
threads.

> - The physical method of message passing is node locale dependent.

??

> Remote node processes serialize objects and use a network protocol,  
> not necessarialy reliable.

same

> - Out-of-band messages are used for Exception Conditions (SRFI-12  
> conditions).

Here we really have something to do with the BALL code base.  Exception
handling is rough & bad, bad, bad ... and in our way already.