Re: alive e/ken style projects?

2015-10-15 9:51 GMT+02:00 Matt Rice <[email protected]>:

> not sure that anybody has tried this sort of thing before, unreliable
> promises seem a bit counter intuitive, though I haven't used promise
> pipelining very much, the impression I get is that modeling
> unreliability makes things weird...
>

Hi Matt,

My PhD thesis largely revolved around marrying E's communicating event
loops, far references and promise abstractions with the unreliable network
conditions imposed by mobile ad hoc networks [1]. My experiments were done
in a language called AmbientTalk [2]. One of the primary differences with E
is that AmbientTalk's far references auto-buffer messages on network
disconnect and will auto-reconnect when the network heals.

This worked out well in our targeted environment because most network
disconnects were assumed to be just temporary (e.g. due to intermittent
wireless connectivity). Obviously this strategy doesn't deal with more
permanent failures. We had various ways of dealing with that, such as
expiration timeouts on the level of individual messages or leasing
strategies on the level of far references. Promises associated with such
messages or references would eventually get broken with an exception if the
network partition remained.

That said, AmbientTalk didn't implement promise pipelining the way E/CapTP
did, so I can't speak to how one would marry such strategies with promise
pipelining.

Cheers,
Tom

[1]
http://soft.vub.ac.be/~tvcutsem/talks/presentations/phd_tom_van_cutsem.pdf
[2] https://code.google.com/p/ambienttalk/

_______________________________________________
cap-talk mailing list
[email protected]
http://www.eros-os.org/mailman/listinfo/cap-talk