VPLS & the Third Mile

We know plenty about the first mile’s access loops and the second mile’s aggregation routers, B-RASs, and DSLAMs. Yet, as consumer triple-play networks get deployed in earnest, the focus will soon begin to shift to the “third mile” – that area of the network in which a mesh of high-speed connectivity is required among the second mile’s aggregation boxes and a variety of content servers, softswitches, and Internet routers.

This third mile of the network today is quite a mishmash of Sonet, ATM, WDM, and PDH (plesiosynchronous digital hierarchy). But given the undeniable trend toward multiservice IP, the third mile is heading inexorably towards one unifying solution: Ethernet.

Edge router vendors take note: Before long, each operator’s third mile will be operated not as a series of point-to-point backhaul links, but as one big virtual private LAN service (VPLS). Whoever offers the winning VPLS solution may end up reaping the largest rewards from service providers selling, not to the enterprise, but to the consumer.

It was always rather inaccurate to refer to the access network as the first mile or the last mile (since the access network often runs to three miles in length), but it caught on for marketing purposes and even made its way into the standards, thanks to the IEEE’s 802.3ah, Ethernet in the First Mile (EFM) spec. That led a number of analysts (including our own) to begin referring to central office aggregation gear as part of a “second mile.” Again, not very accurate, but it stuck, regardless how far that central office actually was from a customer.

I hesitate add to this misnomenclature, but while writing the Heavy Reading report Telco Triple Play: The DSL Imperative, a finding arose I wasn’t even looking for: The network behind the aggregation layer will face tremendous new pressures as the triple-play takes off, forcing some innovative thinking about how to effectively carry packetized voice, video, and data from thousands of customer sites to the carrier core.

The focus in this report was primarily on the first mile – which DSL flavor worked best where, whether DSLAMs or broadband loop carriers would rule this market, and how regional differences would affect the market’s development. Talking with service providers and vendors over the past three months uncovered an increasing interest in looking farther back into the network to understand the implications of moving to a triple-play.

First and foremost, the triple-play represents a flood of IP traffic traversing the access network, upwards of 20-Mbit/s peak traffic per home to accommodate multiple set tops, broadband data voice, and perhaps an HDTV stream. Second, this traffic is truly multiservice, with very different latency, jitter, packet loss, and protection requirements. Third, it is bursty.

Today, ATM-based DSLAMs hand off OC3s or OC12s to an add/drop multiplexer, which carries it up to the edge of the packet core, handing it to a multiservice switch or router. This ATM-over-Sonet/SDH architecture works fairly well for broadband data, as these are both very mature protocols with low-cost interfaces at these rates. Add in IP video, however, and the economics take a turn for the worse. Most DSLAM manufacturers are already building in Gigabit Ethernet uplinks to carry video traffic, and most will be making a wholesale change to IP DSLAMs in the next year, putting all their broadband data and packet voice onto Ethernet for backhaul.

With an Ethernet/IP access architecture in place, the next question to ask is how best to carry it through the metro and regional networks. Because it’s Ethernet, one might be tempted to believe simple Ethernet switches would suffice, but that isn’t the case. Ethernet switches tend to fall down on the three key requirements of the triple-play: scale, quality of service, and protection. This is where VPLS comes in. VPLS as a service delivery layer offers its own triple-play of values to the operator transforming its backhaul network:

  1. Scaleability. Ethernet is quite limited in its ability to scale to the proportions required in a residential triple-play network, where each DSLAM can serve hundreds of subscribers with data, voice, and two or three channels of switched IP video. VLANs simply don’t cut it here over the long term. VPLS, however, has built-in scaleability, taking advantage of MPLS and its ability to support millions of labels.
  2. Quality of Service. Ethernet has no real bandwidth reservation technique that is truly end-to-end, nor any traffic engineering technique optimized for the triple-play. VPLS can guarantee end-to-end QOS and preserve that capability as it scales. Telcos will soon learn that customers know packet loss when they see it and will quickly complain when their video service doesn’t perform. QOS will be essential in retaining customers lured away from cable or satellite.
  3. Protection. Ethernet’s Spanning Tree and Rapid Spanning Tree work well enough for data services but are not up to the challenge when it comes to the mass rollout of IP video services. VPLS uses MPLS-based fast reroute or standby LSPs (label-switched paths) to ensure sub-50ms protection, which is key for both video and VOIP. In many cases, VPLS can be used as the primary protection layer of a network, allowing some operators to build Sonet/SDH-free backhaul networks if their traffic is purely IP-based.
We’ve already witnessed this in action, and all indications are this trend is just getting started. Witness:

VPLS and the triple-play can all be taken as just one more step in the Ethernetization of the network, from the CPE onward. Telco video forces the issue more quickly than any other service today because of its combination of huge bandwidth, massive scale, and mass-market price pressure – three things that have come together for Ethernet in the past, and will continue to drive it forward just about forever.

— Scott Clavenna, Chief Analyst, Heavy Reading

For more on this topic, check out:

For further education, visit the archives of related Light Reading Webinars:

Page 1 / 2   >   >>
paolo.franzoi 12/5/2012 | 1:24:03 AM
re: VPLS & the Third Mile

I think the idea of mixing traffic for the video network with that for other networks is a laudable notion. Laudable but incoherent.

The reason is that it is unlikely that customers will like the channel switch time associated with deep network scanning for channels. That means that broadcast channels will be brought very close to the customer (say the end office).

The implication of this of course is that there is no statistical gain of this across the 3rd mile. So, whats the point of adding a layer 2 or layer 3 aware product? To add cost? To add delay?

You can argue whether Ethernet over SONET/SDH or WDM technologies are better. But content is likely to flow in layer 1 from Head End to End Office.

Scott Clavenna 12/5/2012 | 1:24:01 AM
re: VPLS & the Third Mile seven,

Incoherent probably isn't the right word, but this is a good point. But certainly many operators are already looking at unified IP access architectures, if not deploying them already (KDDI, for example). Even if broadcast channels are brought to the end office via layer 1 transport, there may be an increasing amount of video content that is multicast through the "3rd mile" of the network, relative to the location/distribution of content servers that may not be colocated with DSLAMs.

leafy 12/5/2012 | 1:23:53 AM
re: VPLS & the Third Mile

Incoherent probably isn't the right word, but this is a good point. But certainly many operators are already looking at unified IP access architectures, if not deploying them already (KDDI, for example). Even if broadcast channels are brought to the end office via layer 1 transport, there may be an increasing amount of video content that is multicast through the "3rd mile" of the network, relative to the location/distribution of content servers that may not be colocated with DSLAMs.


Do you mean to say that there will be multicast running across the VPLS domain? What are the ramifications of this? Do many VPLS implementations correctly prune multicast?

Leafy one
indianajones 12/5/2012 | 1:23:53 AM
re: VPLS & the Third Mile I have to agree with Scott here. Content/media servers are typically not co-located at the POP. Content providers are beginning to peer with other carriers such as MSOs at neutral exchanges in some places, but they would rather not if given a choice. Long-haul IP video distribution becomes real in this case.

Also, for the times when video is not using its bandwidth, it can be used by data and other less critical applications.
paolo.franzoi 12/5/2012 | 1:23:51 AM
re: VPLS & the Third Mile
What I am saying, is that there is no VPLS network involved.

To make this work, backhauling at layer one will be done. Doing anything else gives the chance of screwing up video. If there is no oversubscription (statistical gain), then there is no point in introducing a statical multiplexer (Switch/Router). Since traffic must be unblocked, there is no gain.

Scott Clavenna 12/5/2012 | 1:23:38 AM
re: VPLS & the Third Mile Well, there are VPLS networks involved in all the examples I cited, and they are carrying a mix of voice, video and data, and do a fine job of statistical multiplexing, since any on demand video traffic is inherently lumpy (more during the evening, weekends, etc.). Using a layer 2 service delivery mechanism like VPLS lets the operator traffic engineer their networks for the various traffic patterns that the triple play creates.

Layer 1 for broadcast traffic just doesn't hold up in most of these network implementations because it requires too much bandiwdth be delivered too deeply into the access network for no good reason. Multicast is what IP video is all about; I'm suggesting here that a VPLS is an excellent Layer 2 solution to support video multicast, plus provides some additional benefits of scalability and QoS.

In regards to minimizing channel switch delays, that is done at the DSLAM using IGMP snooping (associating members to multicast groups), so the use of VPLS in the third mile has no real impact on channel switching.

This isn't theoretical; it's operational today in some major networks, and indicates a growing trend.

paolo.franzoi 12/5/2012 | 1:23:36 AM
re: VPLS & the Third Mile
You talked about the 3rd mile. Aka the backbone.

Now you are talking about 1st mile. Aka the access.

Multicasting in the access, yep. Multicasting the traffic in the backbone, nope.

Thats my point Scott.

Scott Clavenna 12/5/2012 | 1:23:11 AM
re: VPLS & the Third Mile Well, I think maybe the term third mile is confusing, then, because I envisioned it as not representing the core, but only the feeder/metro of an access network, so the realm of connectivity between/among end offices, POPs, and content hosting locations. In this case, VPLS makes a world of sense and is in operation today in the examples I provided.

A VPLS is created interconnecting DSLAMs with Ethernet uplinks or OLTs with Etehrnet uplinks to sites in the metro/region with content, softswitches, ISP routers, etc. VPLS is an improvement over traditional Ethernet connectivity because of the features of MPLS. Multicasting in this realm does in fact make sense, though I see your point that multicasting across the long haul doesn't always make sense.

paolo.franzoi 12/5/2012 | 1:23:08 AM
re: VPLS & the Third Mile
I think this is true on a metro basis as well Scott.

As far as I can tell, all the US carriers plan to Layer 1 to the End Office. Why? Again, there is no statistical gain for Video Traffic. If you have a 200 channel lineup and a 10,000 line CO do you will still be bringing a GiGE to the End Office for the content. There is no gain.

If we move to Unicast traffic there could be some gain, that would be based upon the way one chooses to engineer the network and the take rate of services. This creates an assumption that there will be a content source in every Metro. This could be true, but is not what I see the US carriers doing. I do see addins for local content and local ad insertions. But I think its cheaper to run a GigE around than build a second head end.

So, what WOULD be carried on this VPLS network? Shrug, you tell me. Traffic between distributed VoD servers?

So my vision of multicasted content is very different than yours. I don't see it growing beyond the GiGE size of pipe. Given that its pretty cheap to pipe around. If you want to make a case perhaps the UNICAST traffic (VoD or Network PVR) can use such a network. I guess so, would be my answer. Not sure what gain you get for it as I think a server in a metro area is going to be quite flat in distribution.

mgillespie 12/5/2012 | 1:22:41 AM
re: VPLS & the Third Mile Well folks, once again. I'm confused.
I agree with points, and I'm baffled by others.
Things have already moved towards ethernet deployment to customers. Within most countries, cost of ethernet in the 'last mile' stretch of the network works out more cost affective when budgeted over three years than a standard [TE]1 circuit - providing within about 3.5Km. Compare the bandwidth between the two and the math is simple. Most companies CapEx over three years, and hell, in three years god knows how cheap things will have become.
This has been evident for years though, and providers such as bredsbandbolaget, and manufacturers such as packetfront have made this clear.
However, I don't see what VPLS and multicast have to do with one another.
Ethernet tin is cheap, and is partly the reason it has been kicking ass in recent years. VPLS & MPLS in it's various forms are great tools for customer aggregation as ethernet by nature is pretty limited. But what are you suggesting here?
Is VPLS coming in to play to keep stuff tidy? Or are you suggesting a profundity far greater, that I have missed?
Page 1 / 2   >   >>
Sign In