re: Extreme Tunes Up Metro StrategyHi Mr. Zippy, There is an assumption here which may or may not be true. Bandwidth is cheap and it can be sacrificed to provide simpler access like Ethernet. That assumption is probably true in some situation and wrong in others..
But, treating this as universal truth and that is where it failed..
re: Extreme Tunes Up Metro StrategyI'm all for ethernet being used more widely, in the metro and long distance areas. I think the fundamental benefit it delivers is lower costs, due to the already commodity and simplicity nature of ethernet technology.
However, one of the fundamental characteristics of ethernet, which are taken advantage of by all popular or soon to be popular protocols eg., IPv4, IPv6 etc., are broadcasts or multicasts ie., the ability to have a node send a single packet, and have the network replicate it to all or a subset of all devices attached to the network.
Reading through this article, it seems all the commentary is saying "get rid of routing, layer 2 forwarding is better and simpler". The problem with this argument is that routing provides two things - fine grained traffic direction based on specific destinations, and broadcast / multicast control.
Anybody who has dealt with a reasonably large, switched ethernet LAN knows that broadcast / multicast traffic can become an issue, as the switches flood this traffic to all nodes, whether they are interested or not.
For multicast, you can implement tricks such as IGMP snooping to minimise this effect, however, to implement that usefully, you need to implement inter-switch protocols to start keeping track of which devices are interested in which multicasts across the whole network. To be honest, I don't know if these "inter-switch" protocols exist - I know them as routing protocols, eg. PIM, MOPSF etc. Oops, looks like you might need to implement routers after all.
So, you can take care of multicasts using IGMP snooping, and multicast routing protocols . What about broadcasts ? Broadcasts still have to be forwarded to all devices, as that is the definition of a "broadcast". How do you limit the scope of a broadcast ? Implement routers, creating multiple broadcast domains.
Radia Perlman has a whole chapter on bridges verses routers ie. layer 2 verses layer 3 forwarding in her book "Interconnections, 2nd Edition".
So, have I somehow missed the point ? Or is it that the attraction of an extremely commodity protocol like ethernet caused people to forget the problems routing is trying to solve, such that they think layer 2 forwarding is the best solution ?
You are not dumb. However, I think that the motivations here are quite transparent. In my opinion, the vendor is accentuating their strengths and downplaying the shortcomings of their approach. The informed consumer is unlikely to be taken in by this little charade and will debit the vendor's reputation accordingly.
Nah, I'm already here, and in a mood. What the heck.
When's the last time Extreme made a strategic move that was anything more than a press release? Where's their WiFi product? Ask the people at Cisco, Trapeze, Aruba or Airspace if they've ever even seen Extreme compete there.
Where are the carrier wins? They gave up.
Where's the 10Gig product? Still at least a year out and won't even compete against Mucho Grande much less Sup720/Constellation 3, Force10.
So what is Extreme actually doing? Squandering their war chest and floundering around. Time for someone to hook this fish and gut it.
re: Extreme Tunes Up Metro StrategySo, have I somehow missed the point ? Or is it that the attraction of an extremely commodity protocol like ethernet caused people to forget the problems routing is trying to solve, such that they think layer 2 forwarding is the best solution ? ----------
Metro ethernet does not necessarily involve pure L2 forwarding in lieu of L3 forwarding. Metro ethernet strategy with MPLS is clearly becoming the choice for deploying in the metro transport. This what the artcile is talking about.
If most of the L3 traffic is aggregated at the central office or the pop(s), I don't need to plop a router all over the metro network. I would be looking for alternatives that are cheaper and easier to manage and maintain. Metro ethernet and MPLS address this situation.
re: Extreme Tunes Up Metro StrategyI am sure I miss something profound in this discussion. But being a novice, I will still ask the question.
I thought MPLS is a sort of Layer2.5. It still depends on OSPF to build the forwarding table and its big advantage is to simplify forwarding and, with the new additions, to provide resiliance and traffic engineering.
That being so, it seems MPLS is rather similar to a L3 protocol than to pure Ethernet.
Would somebody be kind enough to tell me if I got it all wrong and where?
re: Extreme Tunes Up Metro StrategyI thought MPLS is a sort of Layer2.5. It still depends on OSPF to build the forwarding table and its big advantage is to simplify forwarding and, with the new additions, to provide resiliance and traffic engineering.
MPLS is a layer-2 technology for tunneling and label-switching packets. The labels are set up across the network using signaling protocols that work at layer-3. Your creating a layer-2 path across a network (a tunnel) using layer-3 to discover the topology and signal the tunnel setup end to end.
Its big advantage as I see it is that it provides a way of creating tunnels over an IP network in a clean way as compared to IP tunnels. It also provides a way to create tunnels with resource reservations that does not require ATM hardware.
There isn't one overwhelming advantage though. What MPLS does is provide a toolbox that enables many different things to be done that are of use in networks.
re: Extreme Tunes Up Metro StrategyIf most of the L3 traffic is aggregated at the central office or the pop(s), I don't need to plop a router all over the metro network. I would be looking for alternatives that are cheaper and easier to manage and maintain. Metro ethernet and MPLS address this situation.
Is this the "VPLS" - Virtual Private LAN Service - that is being talked about ? I'll admit I haven't done any research into it, so I'm purely speculating. Part of my motivation for asking if "I was dumb" was to see if somebody might give me a quick heads up.
As I understand it, the motiviation for these solutions is to be able to provide the customer with a "routerless" environment ie. the carrier provides an ethernet jack at the customer sites, and all of the sites are made into a single LAN via layer 2 MPLS VPNs across the carrier networks. Any and all "complexity" is in the carrier network, the customer just plugs in the ethernet to a local ethernet switch, and away they go. No routers, no multiple subnets.
Even if bandwith is large and cheap, there are still a few issues that I'm curious about :
(a) what mechanism within the MPLS network is used to ensure broadcasts / multicast / unknown destinations are flooded to all sites that are part of the "LAN" ?
(b) at a certain scaling point, broadcasts / multicasts / unknown destination layer 2 traffic will become excessive, due to either endnodes receiving too much traffic, or the MPLS network struggling to replicate too many packets across too many sites. I think routers would have to be implemented to solve this problem. In that case the "Virtual (single) LAN" would now consist of multiple subnets ie. it wouldn't be a single LAN anymore, which I would think would be the major selling point of this solution to customers.
As I understand it, this solution is basically shifting some of the network complexity (ie. routing-like functionatlity) from the customer site to within the carrier network. I can understand the idea of shifting it, however, the way it is talked about, the complexity seems to have been "eliminated", which is why it is so "wonderful". Even though a lot of experts have worked on the layer 2 vs layer 3 forwarding problem for many years (and I think MPLS is one of the outcomes of that effort), it seems too good to be true if the complexity has just evaporated.
re: Extreme Tunes Up Metro StrategyOne of the ways that this can be addressed is by playing some games with multiple VLANs arriving at the same customer switch. If the provider uses a VLAN to effectively model a point-to-point serial link and then lets the customer switches form those links into a mesh, then all of the multicast/broadcast replication issues remain in the customer switch.
Yes, this shifts some of the complexity back to the customer, but addresses the issues that you raise.
re: Extreme Tunes Up Metro StrategyIs this the "VPLS" - Virtual Private LAN Service - that is being talked about ? I'll admit I haven't done any research into it, so I'm purely speculating. Part of my motivation for asking if "I was dumb" was to see if somebody might give me a quick heads up. ---------------
VPLS is one such service. There are also quite a few applications which can be solved using ethernet private lines. As Tony mentioned earlier in the previous post you can push the task of multicasting/broadcasting/flooding to the customer edge device.
As far your concerns regarding (a) and (b)
(a) One of mechanisms for the MPLS device is to replicate packets on the different LSPs. This is easier said than done. Another mechanism is to use the native multicast mechanism of the underlying layer (if the support exists). For instance in RPR networks, the broadcast/multicast is done using source stripping. Another way to abstract replciating is to create a multicast MPLS LSP.
(b) Most organizations while creating a geographical L2 VPN would balk at creating such a flat network. What you are talking about involves customer sites directly connecting to the service provider's ethernet jack through a L2 switch.
I think the argument is backwards. The customer does connect to the service provider using a router/firewall/NAT and the service provider does not inherit the customers routing and handing off to the POP router. The solution appeals to pure transport service providers or the groups within a ISP which are responsible for transport.
There is an assumption here which may or may not be true. Bandwidth is cheap and it can be sacrificed to provide simpler access like Ethernet. That assumption is probably true in some situation and wrong in others..
But, treating this as universal truth and that is where it failed..
Dreamer