From: MADMAN (dmadlan@qwest.com)
Date: Tue May 11 2004 - 15:09:01 GMT-3
Howard C. Berkowitz wrote:
>>>
>>> In certain, restricted, cases, usually requiring gigabit or faster
>>> transmision media, yes. See such things as Private VLAN Service (more
>>> the IETF term) and QinQ encapsulation (although the latter is not a
>>> WAN protocol, but the distinction between LAN and WAN is less and
>>> less useful). It's also possible using MPLS.
>>
>>
>> Not that restricted actually and on speeds as low as 100M.
>
>
> /* begin Hindenburg news announcement voice
> "Oh, the humanity!"
> /* end announcer voice
>
> Do remember, Sir, that you are speaking to one whose first interactive
> terminal ran at 110 bps. 100 Mbps. Slow. (makes noises about
> impetuosity of youth, then goes into hysterical giggles).
>
> Seriously, I've dealt with the desire to have LANs-across-wide-areas [0]
> for several years, and I still think that while it will work in some
> restricted cases, it's a system collapse waiting to happen.
>
> Broadcasts and multicasts are a significant issue. If the topology is
> simply straight-line, I suppose you can just pass the multicast
> addresses and not worry about replication. That gets more complex in a
> branching tree topology that stil is perfectly valid under spanning
> tree, but where you will HAVE to have a replication function, or at
> least nontrivial forwarding logic that looks inside the tunnels [1], to
> handle multicasts at branching nodes.
>
> While host processor speeds have improved, you still remain vulnerable
> to broadcast storms.
Absolutely but there are plenty of networks that have no more than
100M links that work fine. You have to look at each case individually,
I would not suggest any partciular speed fits all cases.
But hey we are in the bandwidth business, the more the merrrier!!
Dave
>
> Remember, too, when extending what is essentially an L2 service, you
> lose many of the standard tools such as ping, traceroute, and possible
> ARP cache checks. Yes, Cisco is coming up with L2 traceroute
> equivalents, but this is not a trivial problem.
>
> [0] Let me emphasize my concern is with end-to-end use. Using 802.1q over
> dark fiber or even WDM, with a QoS-enabled L2 switch at the customer
> premises connected to the POP is perfectly reasonanle. One can then
> map VLANs to VPNs, and use (G)MPLS in the core. Again, there is L3
> intelligence in such a system.
>
> [1] 802.1q over dark fiber, or even WDM, isn't necessarily tunneled.
> If your scaling technique is QinQ, or other things like L2TP, carrier-
> scale equipment may have to look awfully deeply into the frame.
>
>> We are rolling out a new service that will provide this using 6500s in
>> the core and 3550's at the customer site, MPLS is also being used.
>> The customer can route, trunk or whatever they desire.
>>
>> Dave
>
>
> I know customers _want_ it.
>
> At 2AM last night, I _wanted_ ice cream. I wasn't dressed, my car wasn't
> working, and I'm a diabetic.
>
>>
>>>
>>> Why do you consider this a good idea? It has potential scalability
>>> issues, for example, involving both absolute delay greater than
>>> expected by LAN devices. Another scalability issue is that
>>> uncontrolled growth at both locations may cause the safe number of
>>> devices in a broadcast domain to be exceeded.
>>>
>>> I've been asked to do this many times, on the basis that "layer 2 is
>>> simpler." At a certain level of scaling, it is not. In general, in
>>> the circumstance you describe, I'd route between the sites without
>>> overwhelming evidence to the contrary. I can even keep VLAN numbers
>>> consistent at both sites.
>>>
>>> Unfortunately, studying for the CCIE lab overempasizes lots of things
>>> that may be possible, but not a good idea in the real world.
>>>
> \\
>
-- David Madland CCIE# 2016 Sr. Network Engineer Qwest Communications 612-664-3367"Emotion should reflect reason not guide it"
This archive was generated by hypermail 2.1.4 : Wed Jun 02 2004 - 11:12:09 GMT-3