Re: Socket migration

From: David Brower (dbrower@us.oracle.com)
Date: Sat Feb 26 2000 - 12:35:54 EST


FWIW, the SCO/Tandem "NonStop Cluster" product does do complete
connected socket migration, I believe. This only works inside
the very-tightly coupled cluster, which also allows process
migration across hosts. (Sort of hard to migrate processes
when you can't migrate the connections!). Conversely, keeping
the tcp connection state doesn't do you much good if you can't
move the application state as well, for any but the most trivial
of application protocols. Anyway, SCO/Tandem show that it
can be done for cooperative systems. The question for me
is whether this is particularly useful. It seems easiest
enough to move application programs when between transactions,
and let some sort of application failover re-establish the
connection. Where you really want transparency is in the
case where there is a failure on the server, and you'd like
to be able to failover the application and the connection.
But doing so pretty much requires you to be checkpointing to
independant storage after everything that is significant,
and that seems extremely burdensome and inefficient. So, as
of yet, I'm not particularly sold on the usefullness of
socket migration as now conceived. What would be a compelling
application for that feature?

I think the three-way messaging Alex suggests is very
interesting, and of potentially more use.

ttfn,

-dB

> From: Alex Belits <abelits@phobos.illtel.denver.co.us>
> Date: Fri, 25 Feb 2000 22:59:55 -0800 (PST)
> Subject: Re: Socket migration
>
> On Sat, 26 Feb 2000, a9504915 wrote:
>
> > Hi, is there any information / projects that are about linux socket migration

> I assume, you are talking about connected TCP sockets.
>
> I have never heard about any implementation, but it can be a very
> interesting thing to implement...
>
> ...performance of this thing will unlikely be
> high enough (I assume, migration of sockets is supposed to be used to
> solve performance problems, not to create additional ones).

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:15 EST