On Friday, Aug 8, 2003, at 22:39 Europe/London, Gareth Bryan wrote:

I understand what you're saying, but it does sound quite network chatty.

As cautionary note: I have known customers of commercial app servers
really bitch about the chattiness /between/ nodes in a cluster. This
should be kept to an absolute minimum

I agree. The less communication overhead between nodes, the better.

However, a replicated service doesn't necessarily imply that all requests go over the network. Instead, I'd envisage a local JNDI server on each machine that only communicates deltas in the naming setup to other nodes, so it wouldn't be hit consistently with a bunch of read requests.

This is already possible using a cacheing LDAP server to provide transparent cacheing to a primary LDAP server on the network, for example.

Alex.



Reply via email to