Emerging Technology: SONET Rings Into Mesh -- The Future Of Next-Gen Networks, Part I
networkmagazine.com
Advances in intelligent optical switching prompt a second look at mesh architectures. Will mesh networks bring new flexibility and service options to customers?
by Doug Allen
Network Magazine
04/05/01, 12:11 p.m. ET
In all the uncertainty over next-gen networks and how to make money from them, figuring out how to optimize SONET for all kinds of data remains a constant concern. SONET, after all, was developed in the late 1980s as a Physical layer for the voice traffic of service-provider networks. This service remains predictable and static at 64Kbits/sec, without much ramp-up in use over time. In contrast, post-Internet data service is sporadic, varied, and unpredictable-and it's growing quickly. When it comes to delivering data, one size doesn't fit all.
Providers know customers want greater control over QoS, security, bandwidth, time-of-day service, and protection for certain kinds of traffic. Providing plentiful service options is widely perceived as the carrier's Road to Profitability (cue Celestial Choir). But it's not clear when and how we'll get on that road, or what kinds of services we'll find there.
To glimpse the future, let's start with the present. Today's SONET infrastructure is based largely on rings composed of large circles of fiber with a handful of nodes, interconnections with other rings, and two fiber links. One link carries the working traffic, the other lies dormant until needed in case of an outage. You use this second ring for protection or restoration (see Figure 1).
SONET has many good qualities. Everyone understands how it works, and it's easy to manage and monitor. SONET supports static voice with excellent QoS, as well as fault isolation. And crucially, its restoration time of less than 50 milliseconds (ms) means voice calls aren't dropped if a backhoe cuts a fiber line. Essentially, SONET is a low-maintenance choice, and established carriers love that. Enterprise network managers, in turn, breathe easier knowing their traffic has top-notch protection for mission-critical voice (isn't it all?) and data. Customer faith in the provider's infrastructure is the price of entry for today's carriers.
But these same providers are well aware of SONET's limitations. Its voice-centric design was never meant to support rapid traffic. Thus, bandwidth isn't flexible, both in terms of rates, jumping from 1.5Mbits/sec to 45Mbits/sec, and the amount of bandwidth set aside for protection, up to 50 percent of total ring capacity. Upgrades are costly because increasing bandwidth on one link between nodes requires increasing it throughout the ring, even where it's not needed.
Because each ring has a limited circumference, scaling up also requires many rings to connect over a large area. Speed is limited to the highest SONET rate available (currently OC-192, at 2.5Gbits/ sec); if a system needs more capacity, providers must place independent rings on top of each other, and it's impossible to add bandwidth selectively and enable providers to “pay as they grow.” What's more, each ring functions separately because they can't be managed logically, making provisioning services across rings practically impossible without manual configuration and extensive delays-often as long as three to six months and beyond.
Just Then, On a White Horse-It's the Mesh!
The solution to the preceding problem is the new wave of optical switches offering strong switch and router intelligence, along with a mesh network architecture. A mesh connects every node to every other node in the network, unlike rings which connect nodes only to a small handful of other nodes on that link (see Figure 2). Like most things in telecom, mesh is a great new idea that's been around a long time. In fact, before SONET, mesh was the favored topology for carriers.
“The typical pre-SONET transport network was a mesh network made up of 1 to N linear routes connected together with cross-connects or voice switches,” says Harminder Gill, general manager of global network architecture at 360networks, an international next-gen carrier. “These original mesh networks did not have the common control plane in a ring environment, and could deal with synchronizing, through pointer management, the different delays introduced into traffic that took diverse routes.”
Mesh offers numerous advantages. First, it has no distance limitations because flows are switched along typical fiber runs, not rings. This means switch size constrains capacity, not the fiber's line rate. Without rings, you can deploy increased trunking capacity on just the overloaded routes. In contrast to SONET, mesh networks can be constructed node by node as customer demand warrants, at different speeds (SONET rings must operate at the same line rate throughout the network), and without the need for SONET add/drop muxes at each point.
Likewise, increasing SONET capacity requires deploying expensive overlay rings, which lead to “stacked ring” architectures and require cross-connects or matched nodes, none of which are needed in a mesh scheme. Upgrading node by node should also save the provider money in space, power, and operational and personnel costs (specifics on financials weren't available at press time).
More importantly, carriers can configure meshes using a network-wide control plane, enabling the all-important “point-and-click” provisioning that has grabbed so many recent headlines. Indeed, this is the major knock on SONET-service delivery takes too damn long, both to set up and tear down bandwidth. Customers just don't want to pay for unused bandwidth four days of the week, just so they can have enough to go around on that fifth day for a videoconference or collaborative-sharing support. All carriers at least claim to understand the need to make bandwidth more flexible so customers can turn services on and off like a faucet. SONET, as it's currently defined, can't support this kind of liquid model.
SONET's other disadvantage is that, in terms of protection, it uses greater bandwidth, keeping whole fibers idle for redundancy. But a mesh architecture uses intelligent switching technology to create 1:N restoration. “In a mesh, if you have four routes out of a city, you can split the restoration requirement for one of the routes across the other three routes, as opposed to across only one of the other routes,” says Gill. “This results in only having to maintain 33 percent [1-to-3 mapping, in this case] unused capacity for restoration, as opposed to SONET, where you maintain 100 percent [1-to-1 mapping] unused capacity for restoration.”
Intelligent Optical Switching
Of course, none of this means much without intelligent switching. The new breed of optical equipment is aimed at making SONET more flexible and giving the provider greater control and granularity over a number of service parameters associated with end-user applications, such as QoS, bandwidth, levels of protection, and so on. This gear-from the likes of Sycamore Networks, Corvis, Ciena, Cisco Systems, Nortel Networks, Lucent Technologies, and a whole raft of metro SONET and Dense Wavelength Division Multiplexing (DWDM) players-has driven the optical revolution over the last two to three years.
While many vendors have mesh-compliant products or have promised them, only a small handful have strongly emphasized the mesh network itself, including Ciena, Sycamore Networks, and Astral Point, a metro start-up. Astral Point's Optical Node 5000 supports both ring and mesh architectures, as does Sycamore, but Astral Point focuses more on the metro, while Sycamore's product portfolio extends to the WAN as well (more on these two in a moment). Optical powerhouse Ciena is a strong player in mesh, as attested by its recent $200 million customer win with McLeod USA, a Competitive Local Exchange Carrier (CLEC) deploying a DWDM mesh backbone. Unfortunately, Ciena wasn't available for comment.
These vendors are all working on optical platforms that combine switching, some degree of routing, and key management functionality that gives providers greater flexibility over traffic at the packet and wavelength levels. “Mesh architectures allow carriers to build 'intelligent' optical networks, in so much as each node in the mesh is in communication with each other, and is 'aware' of the network topology and resources,” explains Scott Clavenna, president of Point East Research. “Coupled with the appropriate software, these networks can support rapid provisioning and multiple tiers of restoration that traditional SONET rings can't.”
Carriers see tiered services as the way to better profit margins, so offering multiple grades of protection to both providers and customers is attractive, as neither wants to pay a premium for first-class protection if the traffic isn't mission-critical. SONET offers one grade of protection: 100 percent, using path protection. A mesh offers more variety because it allows for link protection, where a failed link is replaced by a multihop virtual link that joins the original endpoints, and all affected traffic is routed accordingly.
Just how much (or little) protection customers will demand for a given application is unclear, but the idea is to offer a Bronze/Silver/Gold-level service suite, allowing end users to assign a protection class to each application or session. Noninteractive, non-mission-critical traffic can then ride more cheaply than highly confidential executive data, with customers pocketing the savings.
Carriers can also choose to set up protection routes dynamically, either after a failure has occurred or ahead of time for proactive rules setup. A preplanned approach helps maintain QoS.
The downside? “This intelligence comes at the cost of complexity,” says Clavenna. “Managing thousands of millions of lightpaths requires adding significant computing power within optical systems. This isn't a trivial undertaking and will likely take more time than most vendors anticipate.” Chris Nicoll, an analyst with telecom consultancy Current Analysis, agrees. “Meshes require a high degree of intelligence that makes interoperation of networks very difficult. Thus, multivendor network implementations usually require some sort of standards work to provide interoperation at the intelligence level.”
The Best Fit
If we've given you the idea that SONET is no longer extremely valuable, or that it's going away anytime soon, it's simply not true, as any RBOC will tell you. If mesh does overcome the critics, it will coexist with SONET, performing a different mission. Where, then, is each best suited?
“It makes sense to use mesh in any network, CLEC or RBOC, where there are fiber constraints because mesh networks make more efficient use of fiber resources,” says Charlie Wang, Astral Point's senior product manager. More specifically, rings will probably remain at the metro or network edge, while mesh simplifies the core (see Figure 3). The metro has scattered access points and generally cheaper bandwidth, which are best supported by rings.
“The bandwidth reservation issue is less in the metro, where bandwidth is cheaper,” according to Nicoll, “but it's a killer in the WAN, where bandwidths are much higher. In the metro, you may have to reserve bandwidth on an OC-48 ring, but in the WAN, try justifying having to reserve 60 OC-192 wavelengths.”
Traffic type and access characteristics are significant too. Rings are better suited to time-sensitive flows (VoIP, videoconferencing) across a relatively small area, such as in a metro or state. Mesh is more appropriate for non-real-time data moving across larger geographies, such as cross-country or internationally. “If traffic demand is high and unbalanced across routes, mesh becomes attractive, even in a small geography,” says Gill.
Metro mesh might be more difficult since the fiber plant is not deployed with mesh in mind. As Clavenna points out, most customers reside on access rings, not points on a mesh. According to Tom Nolle, president of CIMI Corp., “SONET works in fairly thin-demand applications, where mesh architectures only work where there are enough nodes to build a fail-safe mesh from. That tends to make SONET work better in access applications, but less well in interoffice applications. And SONET is compatible with the physical architecture of RBOC fiber, which is wired as a 'star of stars.' Mesh architectures may require strands where none go.”
But in the core, a mesh should make it easier for capacity wholesalers to offer differentiated services to retail customers. Optical circuits can be sold at various levels of restoration, with much quicker provisioning. |
