On Fri, 2002-08-09 at 11:22, Hanson Char wrote: > My assumptions are based on the use of either TCP or UDP transport > protocols. > > For UDP, the broadcasting is not reliable, hence cache items will gradually > get out-of-syn. > > For TCP, a peer-to-peer broadcasting model means n(n+1) TCP connections > among the n nodes, versus 2(n-1) in the case of using RCS. Hence it's just > too expensive. > > The catch-22 of either being too expensive or too unreliable leads me to my > (potentially incomplete) conclusion.
Ahhh, okay. I agree with both conclusions actually, but building on top of javagroups addresses these concerns I believe. By using multicast UDP we can eliminate the n(n+1) issue. Then, by implementing a reliable retransmit protocol on top of that (which javagroups does already) we get reliability. It's a really clean solution IMHO. Javagroups provide so much. It can be configured to simulate multicast with unicast if required, or to traverse firewalls, or any number of other things... I'm still testing, but I'm planning on moving to it from the TCP lateral soon. -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
