I first posted this mail in the Sequoia mailing list, but I thought it would
be better to post it here.

====

Hi there,

Just a little post to tell you I got an alpha version of a pure JMS
implementation of Hedera working.

For the time being, I only tested it using the supplied
ChannelPerformanceTest and the performance level of this implementation is
quite as good as the one performed by the current JGroups implementation. I
used a single instance of SwiftMQ 5.2.5 as JMS Server with two Hedera nodes
connected to it.

The total ordering seems to be respected too so all I need to do now is to
test it in real world scenarios with Sequoia.

The only problem for which I haven't found a satisfying solution is the
following : 
- in Hedera each member (node) is identified with an IP address, a port and
an identifier
- using JGroups this is OK because the combination of IP address/Port is
unique for each node and the current implementation set the identifier to be
equal to the group identifier.
- using JMS, you know the JMS server IP/Port and the client IP address
(localhost) but you have no way to determine the client port !

This is a problem when you run two nodes on the same server because the IP
address and the identifier are the same. As a work-around, I compute a
random (dummy) port number for the local member.

This is not a major problem since running two nodes on the same server is
not the recommended architecture to use with Sequoia but I'll try to find a
better solution anyway.

Best regards,
Karl Cassaigne




_______________________________________________
Hedera mailing list
[EMAIL PROTECTED]
https://forge.continuent.org/mailman/listinfo/hedera

Reply via email to