Subject Re: [IB-Architect] Messaging API
Author Jan Mikkelsen
Jim Starkey <jas@...> wrote:

>Whether or not any of this leads to anything useful is interesting
>but moot; our job is to define appropriate services in our problem
>space with the existing technical infrastructure.

Let's recap:

My original message proposed a protocol which used an unreliable network
multicast service to provide a general purpose reliable message multicast
service. I proposed using IP multicast as one possible network service
because it could cross router boundaries and wasn't restricted to a single
LAN. I also stated that the general mechanism could use LAN multicast
mechanisms or other multicast services.

Your response was to say that broadcast couldn't work because it was
restricted to a single LAN, and therefore wouldn't work on the "Internet".

My response was to restate what I said in my original message, that IP
multicast worked across routers and used WAN links. I gave a few
references. I also proposed a fallback mechanism for supporting particular
clients whose network operators did not support IP multicast (essentially
falling back to your proposal should using a multicast technology fail
consistently for a given client).

I'm clearly missing something: How, exactly, is this not working within the
technical infrastructure?

In any case, let me add a point:

I believe that the majority of users of this functionality will be within a
network under the control of a single organisation. Assuming this, the
addressing issues of IP multicast can be controlled, and the network can be
sized to ensure that broadcast packets are rarely dropped. This is
possible. I know: I've done it.

Incidentally, I don't think you've shown that your proposal will scale well
even in controlled network environments, let alone over usage tariffed WAN
links, assuming many clients are interested in the same messages. If there
isn't a significant overlap between client interests, of course it doesn't
matter much.

I think this has been pretty much discussed to death. We're just restating
arguments now. It's all a bit pointless without the source and the ability
to implement, anyway.

Jan Mikkelsen