On Mon, 10 Jan 2011 14:45:05 +1300, Brian E Carpenter
<[email protected]> wrote:
Fred,
I'm confused. We've been talking for months about recommending
pseudo-random flow label values as inputs to hash functions,
precisely to allow scaleable and stateless load balancing and ECMP.
I completely agree that per-flow state doesn't scale.
Regards
Brian Carpenter
On 2011-01-10 14:26, Fred Baker wrote:
The issue is that randomness doesn't help, if you want a scalable
approach.
It means that for each flow passing through the load balancing
system, I have
to store its flow label and assign it a path. What are the arguments
against
NATs? One of the big ones is the expectation of per-flow state in
the network,
isn't it? If you're expecting the network to store per-flow state,
you by
definition have a scaling problem in the network.
To use the flow label in load sharing and have it be remotely
scalable,
the network needs to be in control of the label.
Perhaps Fred has applications of "load balancing" beyond ECMP or LAG in
mind?
The network doesn't control port numbers, so his argument obviously
doesn't
apply to ECMP or LAG.
Regards,
// Steve
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
