Hi,
Just getting caught up on the HA list...
Alan Robertson wrote:
Lars Marowsky-Bree wrote:
On 2005-11-11T16:23:04, Alan Robertson <[EMAIL PROTECTED]> wrote:
This is something which the CGL people had asked for in the past - so
we included it. It was on our schedule for 2005. Perhaps it will be
useful, but perhaps not.
The protocol is the only one I know of which is specifically designed
for clusters. It includes reliable multicast, etc.
The direction is to be useful. This might be useful to some people,
and if they do the flow control and retransmission in the kernel
correctly, then heartbeat can benefit from that.
Not sure what you mean exactly.
All messages are delivered reliably to a node but there
is no flow control at the socket level
(at least for message oriented sockets)
This can either result in returned or dropped messages
depending on the sender socket type created by the user.
...
Right, no disagreement at all ;-) I like having the support for it in
the project, I was just trying to understand whether
a) it provides benefit over what we already have,
Heartbeat between all nodes - in kernel.
User-space would subscribe to a node join/leave event.
and subsequently b) because if it does, I'd look at putting TIPC into
SLES.
Cool.
TIPC is in linux 2.6.16-rc1 so I assume that distros will ship with
it once it becomes a fully qualified release - right?
If I get your reply correctly, right now it doesn't provide anything to
us which we don't already have?
My (limitted) understanding so far is that TIPC, unlike for example
openais, doesn't have it's own notion of node liveness, and so it would
just be an other communication channel. I know it provides some
redundancy, but we also already have that in our own code base...
Well, TIPC does have a virtual link between all pairs of nodes
and an api for user code to be informed of changes to this topology.
I'm just trying to extort more information about TIPC from the list
members so I don't have to go read up on it myself, in the hope that
someone else has already done so ;-)
I know a thing or two about TIPC, ask away.
Supposedly when in multicast mode it can/does provide reliable transport
- and there's no way to do that without node liveness information.
I'm not familiar with the mcast implementation but I'll check into it.
If it also does its own flow control, then that's also a good thing.
yes, with certain restrictions as noted above.
--
// Randy MacLeod
[EMAIL PROTECTED] | Not Nortel's Opinions.
---------------------X-----------------------------
Nortel, Ottawa, ON / \
_______________________________________________________
Linux-HA-Dev: [email protected]
http://lists.linux-ha.org/mailman/listinfo/linux-ha-dev
Home Page: http://linux-ha.org/
