river-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sean Landis" <sean.lan...@gmail.com>
Subject Re: JavaSpace.notify() "not reliable"
Date Fri, 18 May 2007 14:56:01 GMT
I've heard a few people say they cannot use Outrigger (and I think
Blitz too, but I'm not sure here) because of the excessively high
failure rate of notify(). I'm afraid I haven't done the testing
myself, but those who say this tell me that the failures are more
related to the implementation than to the typical reliability
challenges.

Jini/JavaSpaces are designed to help the developer face the realities
of distributed computing (8 fallacies), but still the software should
do a really good job when all other things are working as hoped. It
shouldn't be an additional source of challenges.

Regarding polling, that seems counter-productive. I understand Gregg's
arguments about truly reliable software, but it seems that we aren't
just talking about dealing with unreliable networks but rather a
workaround for the additional unreliable nature of the notify()
implementation.

I realize the implementation in question conforms to the
specification. That doesn't resolve the issue.

Does anyone see this as an impediment to adoption? Should the notify()
implementation be visited going forward?

Sean

Mime
View raw message