[ticker-dev] presence proxy agent

[David Arnold]

-->"Michael" == Michael Lawley <lawley at dstc.edu.au> writes:

  Michael> y'know, a proper persistent tuple space (linda :-) would be
  Michael> just the ticket here.

maybe.  linda is a fine model.

 from an implementation point of view (and this is probably more
elvin-dev fodder), Elvin made two major simplifications from the Linda
model:

1) the Linda in() operator requires tuplespace-wide synchronisation.
   Elvin effectively only implements out(), and rd(), removing this
   difficulty.

2) Linda's storage requirements mean that in practice the model is
   diluted by engineering concerns for management of storage.
   publishing Usenet into pure-Linda is not feasible, for example.

these choices make wide-area Elvin far more practical.

additionally, they have pushed us more into a model of prolific
publication which i believe has intrinsic merit as a way of
constructing systems compared with Linda's more coupled approach.


having said that, there're obviously uses for persistent state.
persistence in a data-oriented model like Linda is simple.  Elvin is
more challenging: it is both message-oriented (leading to mailboxes)
and data-oriented (leading to ...?)


it has long been my hope/belief that correlation is the real way to do
persistence in Elvin.  in the meantime, i guess we can keep playing
with simpler approaches ...




d