Stepping back one notch from this discussion, there's a basic
architectural point here.
As Milo Medin is fond of saying, "with enough thrust anything will fly"
and sometimes they get enough thrust to semi-permanently embed themselves
into the networking infrastructure.
ATM is a case in point. By about 1993, it was clearly losing steam, but
so much money had been put into it, that it was already part of the
infrastructure, and once in, it wasn't leaving soon. (Side note,
the marginal cost of using an inferior technology that is already
installed is often lower than the cost of installing the better technology).
MPLS has a genealogy that leaves it suspect (it descends from a vendor
response to IP switching -- and IP switching turned out to be a fad) but
a lot of careful work has gone into trying to make MPLS a sturdy technology.
The issue is, has that work succeeded?
I'm actually not in a good place to say. I know some of the things people
say about MPLS are clearly silly (the notion MPLS is faster to switch than
IP reflects a poor knowledge of router innards, or a poor router design).
Other statements have some credibility -- carriers have long wanted to do
overlay networks to better track resources (witness how UUNET ran their
backbone a few years ago) and MPLS apparently can help.
Craig
Unnamed Administration Sources Report that Criag Partidge said...
{snip}
MPLS has a genealogy that leaves it suspect (it descends from a vendor
response to IP switching -- and IP switching turned out to be a fad) but
a lot of careful work has gone into trying to make MPLS a sturdy
technology.
The issue is, has that work succeeded?
I'm actually not in a good place to say. I know some of the things people
say about MPLS are clearly silly (the notion MPLS is faster to switch than
IP reflects a poor knowledge of router innards, or a poor router design).
Other statements have some credibility -- carriers have long wanted to do
overlay networks to better track resources (witness how UUNET ran their
backbone a few years ago) and MPLS apparently can help.
The scary thing is that the "speed" of MPLS-based networks is taken as
gospel by an alarming number of engineers, mainly those who have come out of
the large telco's (i.e. ILECs), and are still kind of mad that ATM didn't
work out. These folks are more or less alarmed by IP, and desperately seek a
more deterministic, switch-based model of data transmission for the Internet
as a whole. The fact that there is no practical, real-world difference in
forwarding speed between straight IP, and IP over MPLS is generally
explained away by these guys in a fairly elaborate handwaving exercise. At
least one major hardware vendor is not helping this, with some of their
engineers convincing major customers that conventional IP routing is bad,
and that anything MPLS is good. While I agree that MPLS has it's uses - i.e.
TE as an exception handling mechanism for outages, and L2VPN technology as a
FR/ATM replacement, some folks need to approach the technology with
additional caution, and not blindly embrace it as a panacea. As the internet
engineering community evolves, learning from things like ATM, becomes quite
important.
- Daniel Golding
Craig Partridge wrote:
the notion MPLS is faster to switch than IP reflects a poor
knowledge of router innards, or a poor router design.
Or just outdated information. It wasn't too long ago that IP best-
match lookup hardware couldn't do line-rate at the high speeds needed in
the cores of large networks. At that time, ATM was able to do line-
rate forwarding at those speeds. This is one of the reasons that ATM
was adopted in those networks. One of the big reasons for ATM's speed
is its use of fixed-length VPI/VCI header lookups instead of variable-
length IP best-match lookups. MPLS, thanks to its use of a fixed-
length label header lookup has the same advantage.
Over time, however, this advantage disappeared. IP best-match lookup
chips were developed that could do a proper IP lookup at full line rate
for OC-48 and even OC-192. With line-rate IP lookup, it's no longer
possible for something else to be faster.
It is possible that the pendulum may swing back the other way in time,
of coruse. It is possible that IP lookup chips that can handle the next
generation of line rate may not become available in a timely manner,
which would once again give the advantage to fixed-length header lookup
chips. But not having a crystal ball at hand, I have no idea if this is
actually going to happen or not.
-- David