Re: Comments on the L2 VPN framework and solutions documents

Eric Rosen wrote:
> Mick> Of course the IEEE model doesn't require all provider bridges to be on
> Mick> the  same emulated  LAN,  only  all those  that  are optimising  their
> Mick> control traffic by  sharing its results between them  - i.e. just like
> Mick> ordinary LAN. So VPLS's that are on the *same* underlying connectivity
> Mick> can  share  that knowledge  and  don't  have  to suffer  BPDU  sending
> Mick> inefficiency. 
> 
> So the difference between  the IEEE model and the FW model  is that the IEEE
> model allows this optimization.  But I don't quite understand how to benefit
> from this optimization in the  VPLS environment.  A particular set of VPLSes
> could only be considered to have the same "underlying connectivity" if it is
> known that they  are supported by the same  set of PEs, both now  and in the
> future.  I'm not  sure whether this is a common enough  case to worry about,
> and  I'm  also  not  sure  how  one would  get  this  information  from  the
> auto-discovery provisioning model. 
> 
> 

The benefit, at least to me, is that everything else can treat the VPLS 
instance as a LAN and doesn't need to be aware.  So provisioning and 
forwarding behavior for bridges doesn't need VPLS awareness.  Otherwise, 
  certain higher level functions have to know that a port is a VPLS port 
or an Ethernet port.

Additionally, it allows arbitrary topologies to be built.  Right now, 
most people are on the direction that there's a single VPLS per customer 
and that it spans edge-to-edge.  And people are complaining about the 
scaleability of that solution.

Back in older days with previous incarnations of emulated LANs, many 
people designed their networks differently.  I recollect some carriers 
building mostly regional emulated LANs, putting multiple customers 
within a region on the same regional VPLS and putting bridges together 
between the regions.

The logic as I recall was that bandwidth within a region was pretty much 
free, but between regions it was real money.  Many customers stayed 
within a single region, which was even better, and multicast/broadcast 
traffic was policed properly.  The specific example I recall was a 
carrier that had 5-10 switches in NY and DC, and they created a single 
emulated LAN instance in NY and DC with VLANs across it, and they had a 
bridge between the two that switched a small number of VLANs between NY 
& DC.  Later, they replaced the emulated LAN with a big honking Ethernet 
switch and everything continued to operate.

Now I'm not promoting this topology over any other.  But ensuring the 
emulated and real technologies are transparent to the higher layer, one 
can build arbitrary networks based on traditional bridging models.  I 
hope thats a good thing.

- Matt