[Uni-verse] Report on the state of UVAS

[Peter Lundén]

We have found the cause of the back-end protocol problem. Its a rather
weard problem, we found that UVAS is trying to write to fast to the out
going port. You might believe the system call would block until it can
handle anther call but it does not, infact trying to call it again to
quickly might destroy the former packet and that is what happens. Some
of the packets never leaves the computer.

Raine came up with a simple fix which was to delay the call by adding a
printf line right after it. However this caused a new problem, UVAS was
unacceptably slowed down by the message reply mechanism. To get around
this we changed the object allocation interface so it now allocates
local id number in each peer, the id number of the client is sent over
to the server and then the server set up a mapping between the id numbers.

The current status of the protocol is that we believe it is stable
enough for the demo. The sending mechanism in protocol should however be
reimplemented using message packing, that will probably resolve the
problem. If I get any time I will do this before the demo but it has low
priority right now.

The new geometry handling in UVAS is implemented and is working except
for some bugs. It was a much harder problem then a though from the
beginning. There are still some bugs to fix in the updating mechanism.
and some tuning is needed to get the new geometry handling work well
with the updating. I dont think there are any major problems in UVAS
now, just a couple of bugs to fix. My plan is to try to do some more
testing against Samuels reducer early next week.

Im confident that we will have a nice demo running at the IST conference.


--PLu