Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

[EFM] What Ethernet are we talking about?




Andrew,

> That doesn't sound very much like the Ethernet or 802.3 that this
> committee has worked on for the last few decades. Are you sure
> this is the right place to get your requirements met?

First of all, I need to apologize not only to you, but also to the group. I
believe that I've been somewhat avid to hit the 'reply' button, and some of
my comments reflect this lack of patience. So sorry if my comments seem a
little out of place sometimes.

Now for the real reason behind my comments about EFM requirements. I
believe that the recent experience that most people (including me) have
with newer Ethernet equipment may have biased our perception of what is
Ethernet about. I've been a user of Ethernet for 15 years or so. It's not a
long time for telecom operators, and even not a long time for Ethernet
itself. However, given the evolution that we've been experiencing in LANs,
its a very long time indeed. I've been using half-duplex Ethernets on coax
cabling (remember the NE-2000 days?) years before Win3.11 made TCP/IP and
Internet readily available for MS users, so I've been through parts of this
long history.

Ethernets of today are mostly switch-based P2P networks. Collisions are no
more, and VLANs allow us to effectively isolate the broadcasts. In the end,
the user experience is completely different from the one before, where we
had a collision-based, shared-media network; these limitations gave way to
a high speed, full duplex switched network.

This change of perspective has a lot to do with my comments. It proved that
we can have a excellent mix of the original Ethernet and point-to-point
technology. The switched Ethernet of today is able to offer almost the same
services that a Frame Relay network offered a few years ago, with much
higher speed and much lower cost. As a benefit, you still get the VLAN
concept, which is a much better implementation of the point-to-multipoint
access network than any of the Frame Relay based alternatives, and makes
scaling multicast applications much easier.

The proposal to offer Ethernet-based access over a PON put us back on some
of the questions that we were facing about 10 years ago, when modern
Ethernet switches first arrived. On PONs, the shared media is back in the
game. Some newer standards, such as 802.1q, work beautifully on switched
LANs. However, the compromises that were made to make it compatible with
older switches make the implementation of VLANs over PONs tricky at least.
For instance, if some user puts a EPON compliant bridge on the segment,
what does stop him from seeing all the broadcasts, even the ones from
different users or VLANs? And how do you encrypt traffic per user, or per
VLAN? Of course all these concerns can be (and they will be) handled; we
could go and discuss the details, but that's not my goal, at least now. I
just want to make it clear that we have reasons to raise these concerns,
and I would like to see them openly discussed in this group.

My last comment may seem a little out 'off topic', but it is a big concern
of mine. Ethernet is only the layer 2, and it has to make a nice job at
this. Please don't try to make it a single solution for all the problems.
I'm yet to see someone collapse the entire network in a single 'fat layer',
mixing L1, L2 and L3 functionality together, and make it *scale*. That's
where the problem lies. This is part of the reason why I'm so concerned
about VLANs and point-to-point connectivity; we need to have flexibility to
be able to map any logical L3/IP topology over the basic L2/Ethernet access
network. Limitations on the L2 topology will take out a lot of the value of
the EFM proposal.


Carlos Ribeiro
CTBC Telecom