On 31/12/2013 20:48, John C Klensin wrote:
For example, effective surveillance of traffic content by monitoring
of Internet links would become considerably more difficult if we
(pervasively) went back to routing at the packet (or datagram) level
and optimized our routing algorithms to prefer diverse paths within a
stream or flow. At least as I understand it, that would largely
eliminate the use of MPLS and would slow things down overall unless
ISPs started engineering their networks for more path diversity
between any two endpoints (presumably increasing costs for the amount
of traffic handled). But it would make interception of a single flow
for surveillance purposes a lot more unpleasant and costly for the
monitoring body without requiring encryption.
Actually John, I think some MPLS implementations did this across equal
cost paths, but TCP performance went through the floor and so a lot of
effort now goes into the maintenance of flow order. Some protocols are
much more sensitive than TCP in this regard and can never be subjected
to misordering. However if the transport protocols could cope with out
of order delivery, we could get MPLS to spread flows across pretty much
every available path. As well as having anti-surveillance properties,
such an approach would have the anti-congestion properties of spread
spectrum and so could be quite an interesting area of research.
Stewart