"The Role of Optical Internets in the New Public Network"
==============================
All,
SilkRoad, and others entering the bandwidth race with innovative approaches, must find ways to introduce their new technologies to carriers' staffs in such a way as to meld with their existing behavioral and adaptive processes. It wont be easy.
WDM, and then DWDM, were first viewed by Service Provider engineers and Operations Staff in the most awkward of ways, just a few years ago. But now these techs have become a mainstay that will be difficult to dislodge. The DWDM course seems to have been set by the architects, and this article addresses where that course leads.
The following article depicts the DWDM model of Packet over SONET [POS] and Packet over Lambda. The startups technologies, such as those being introduced by SR and others, must demonstrate that they will surpass this model in both resource efficiency and overall payload capacity.
What's interesting about this article is that it also demonstrates that anything new out of the box must also be prepared to handle routing and situations calling for robustness, lest they be relegated to point-to-point solutions only.
It's from the January 1999 issue of Telecommunications Magazine, best viewed at:
telecoms-mag.com
The article is reprinted below, minus the topology diagrams, for thread posterity.
Enjoy, and Regards, Frank Coluccio
=====================================================
"The Role of Optical Internets in the New Public Network"
by Andrew K. Bjerring and Bill St. Arnaud
A new network architecture based on IP over DWDM promises to lower costs and reduce management complexity by eliminating the ATM and SONET layers. Is this a viable model?
Andrew K. Bjerring and Bill St. Arnaud
------------------------------------------------------------------------
There is considerable debate in the industry about the best technology for transporting IP services, with the basic choices depending on whether or not ATM and SONET technologies are layered between IP and dense wave division multiplexing (DWDM). In the short run, the top consideration is the extensive deployment of networks using those middle layers and of services based on them. Given the dramatic growth of internet traffic and the development of new internet applications, however, just about every kind of traffic will ride, in time, over IP. A network optimized to carry the unique properties and capabilities of IP traffic, such as an optical internet, will clearly result in dramatic savings.
Recently, plans to deploy optical internets, which are often referred to as “all IP,” or IP/DWDM networks, have been announced by Sprint, Frontier, Global Center, Enron, and the Canadian Network for the Advancement of Research, Industry, and Education (CANARIE). The defining characteristic of optical internets is that the network link layer connections are dedicated wavelengths on a DWDM optical fiber directly connected to a high-performance network router. In such a configuration, the router replaces traditional ATM and SONET/SDH switching and multiplexing equipment.
The primary rationale for such networks is the significant reduction in capital and operating costs as compared to a traditional packet over SONET (POS) network, or even an IP/ATM/SONET network. Estimates of the capital cost savings range from 50 percent to as much as 90 percent compared to a traditional layered internet. Moreover, through the elimination of the SONET and ATM layers, the operating cost of an optical internet can be reduced by up to 60 percent. Of course, savings of this magnitude are generally only possible in a new network deployment.
Telcos have frequently argued that an ATM/SONET infrastructure will be essential in making the Internet as reliable as the telephone, particularly as the volume of data traffic grows. However, the same degree of reliability can be achieved at IP Layer 3 without an ATM or SONET layer. The only reason IP rides on top of ATM and SONET is because IP traffic is small in volume and has to be muxed with other ATM or time division multiplexing (TDM) services for cost-effective delivery. The Internet, from the very early days, had the capability to bypass the SONET and ATM layers. In fact, it is currently estimated that only about half of all internet links use ATM/SONET circuits, with the bulk of the Internet traffic being carried on unprotected leased lines.
The DWDM Revolution
The dramatic economic advantage of an optical internet is based on several factors, the most obvious being the nature of DWDM itself. Telecom equipment vendors have announced a plethora of DWDM products over the past two years. Channel capacities on these products have rapidly increased from two and four channels up to 16. Recently, systems with more than 100 channels have been announced. These developments alone represent a dramatic increase in the bandwidth capacity of installed fiber. DWDM products are rapidly being deployed in long-haul backbones, where economics dictate the largest advantage will be gained and where new construction presents the first opportunities for deployment.
Recently, however, companies like Ciena, Nortel, and Cambrian Systems have announced DWDM solutions that address the economics of the metro network as well. Unlike long-haul DWDM, these systems do not have to be designed to squeeze the maximum number of bits possible through a given fiber, since bandwidth efficiency is less of an issue at the metro level. Metro DWDM solutions tend, therefore, to be considerably cheaper. The ultimate impact of this could be that metro DWDM will have as great an impact on the economics of the local loop as DWDM is already having on long-haul economics.
Traditionally, carriers multiplex different network services into a single transport stream, most commonly using SONET/SDH TDM. More recently, ATM has taken on a larger role, particularly in the broadband carrier market, or wherever a cell- or packet-based multiplexing schema is needed.
The advent of DWDM networks is changing all that. It is now possible to have optical multiplexing, with different wavelengths being used to support different network services. For example, on a single DWDM system, some wavelengths might be dedicated to a high-bandwidth, optical IP network, while others might be dedicated to an optical ATM network. Still others would be dedicated to traditional SONET/SDH services, including the support of overlay network protocols such as IP and ATM.
Unique Characteristics of Internet Traffic
One of the recent discoveries about the Internet is that traffic patterns on internet networks look the same regardless of the number of simultaneous sessions. Conversely, traditional traffic aggregates with the number of users following well-understood patterns: random calling patterns on individual lines aggregate through local exchanges and inter-city loops up to national backbones that show relatively smooth changes in total traffic volumes over time. Traditional network planning is based on well-established queuing models for predicting traffic loads.
Voice network traffic aggregation models work for two simple reasons: voice traffic aggregates in 64-kbps steps with each phone call and once a link is saturated, no more callers are admitted to that link. Since neither of these statements applies to internet traffic, the need for a fundamentally different approach becomes apparent.
On the Internet, any single computer can effectively use as much bandwidth as is available, although it will normally do so in random bursts of activity. There is no practical limit to the number of computers that can access the network at any given time. Finally, since there are no busy signals, a computer experiences traffic congestion in a very different way than does a voice caller. In the event of congestion, all computers using the congested link back off and begin retransmitting at a lower rate until the congestion clears. This pattern of saturation followed by back-off results in waves of data being transmitted on most internet links. It is this combination of the power of each computer to generate traffic and the way in which congestion is handled that produces “bursty” traffic at even the backbone level, which is what underlies the fractal model of internet aggregation.
A common solution network operators use to accommodate bursty traffic is to install large buffers on routers at network ingress points to smooth out the peaks and valleys. However, large buffers introduce jitter and throughput delays, which are unacceptable for some applications, particularly real-time voice and video. An optical internet provides an alternate way to handle bursts by using the protection bandwidth that sits idle in most SONET networks.
Most modern fiber systems are deployed in ring architectures. In a traditional SONET network, one side of the ring is used as a backup, or protection circuit, for the other working side of the fiber ring. In the event of a disruption of service to the working fiber, the traffic is automatically switched to the protection fiber (see Figure 1). For many internet applications, however, that level of protection is not needed, since computers handle disruptions in service through retention of packets at the source.
In an optical internet, therefore, the routers can effectively make use of the wavelengths on both the working and protection fibers in the ring (see Figure 2). The effect is that the protection fiber becomes extra bandwidth to handle the large data bursts associated with internet traffic, which can then be transmitted without buffering in the router itself. In the event of a cut in one side of the ring, the router would be able to revert to more traditional buffering and flow control mechanisms.
[Picture: Fig. 1 Traditional Internet Architecture]
Another unusual characteristic of internet traffic is the extreme imbalance that exists between the transmit traffic flow (Tx) and the receive traffic flow (Rx) on most links. This asymmetry is due to the presence of large server farms that generate large amounts of data, such as Web pages, in response to a simple request from the user.
An optical internet can easily be configured to take advantage of these asymmetric data flows. By coupling routers to individual wavelengths, the direction and number of wavelengths can be configured to closely match the traffic load on any given link. The typical asymmetry ratios on major internet links vary from as low as 3:2 to as high as 16:1 at some large internet exchange points. Optimizing the allocation and direction of DWDM wavelengths to match this asymmetric traffic flow can result in a cost saving over traditional balanced and symmetric SONET links.
Basically, traditional networks have been designed for two-way conversations, and hence all circuits have been automatically configured as bi-directional paths. The Internet, on the other hand, was designed on the assumption that Tx and Rx data flows are independent. Currently, these independent, uni-directional paths tend to be configured on bi-directional trunks through interface cards on the routers, but this is not required and generally doesn't make sense. Not only could the router be configured with distinct Tx and Rx paths, but it could also be configured to support multiple parallel channels along the same physical link. By optimizing the direction and number of wavelengths to match the offered traffic load, up to a 50-percent savings could be realized.
Goodbye, SONET
One obvious source of savings in an optical internet comes from the elimination of the SONET layer. This step raises more fundamental challenges for the network engineer, including survivability.
As modern network overlays become increasingly robust, the fault tolerance built into the core transport network is becoming increasingly redundant, particularly for IP networks. In short, efficiencies and cost savings can be gained from single layer management of survivability.
One of the major advantages of SONET ring networks is that they have been able to do restoral in as little as 50 milliseconds. In an internet, the timers on router interfaces can be configured so that they will switch or provide path failure notification just as fast. In theory, a router could be placed at the end of a fiber link instead of at the end of a SONET terminal to provide automatic protection switching.
However, although individual routers can be configured to respond quickly in the event of a fiber cut or node failure, the route table update required because of the failure must be propagated to adjacent routers as well. That propagation can take several minutes or several hours, depending on the configuration of the network.
However, a new protocol called Multi-Protocol Label Switching (MPLS) promises to eliminate the need to propagate routing updates in the event of a local network outage. Once the standard is approved, MPLS will allow optical internets to carry out restoral and path protection switching at Layer 3 (the IP layer), rather than Layer 1.
[Picture: Fig. 2 Optical Internet Architecture]
This offers a considerable degree of flexibility in terms of multiple restoral paths and provisioning of differentiated restoral levels of service based on the requirements of the application or the service level agreement with the customer. For example, a Layer 3 restoral service could offer a different restoral capability for voice traffic than for TCP/IP data traffic, which is normally less critical because the host computer keeps all TCP/IP packets in memory until it receives confirmation of their receipt.
Lower Cost Tributary Services
The elimination of SONET multiplexers, commonly referred to as add-drop multiplexers (ADMs), is another potential cost savings. This has different implications at the backbone and tributary levels. As a rule of thumb, a SONET ADM on a backbone network costs close to $1 million. At the backbone level, an optical add drop multiplexer (OADM) will make it significantly easier and cheaper to drop a complete OC-48 or OC-192 channel from a DWDM system. Downstream, however, a different issue arises. Currently, when a carrier drops off an OC-x to a customer, it breaks out the circuit at the SONET mux in the central office (CO) and then backhauls a tributary circuit to the customer. In general, carriers do not bring an entire OC-192 SONET transport to the customer premises to drop off an OC-x tributary circuit.
With an optical internet, however, particularly in situations where the fiber is running past the customer premises anyway, it is just as easy to break out the tributary circuit locally, rather than backhauling it from the closest CO. This can make a major difference in long-distance backhaul costs, such as when a tributary circuit has to be dropped at small cities and towns along a major fiber route.
What makes this new approach feasible is the equipment needed at the customer premises to break out the OC-x circuit. For example, if the WDM channel uses POS framing, then a high performance POS router or a SONET ADM would be required to break down the WDM channel into multiple access services at OC-x rates; this is most feasible at the CO. On the other hand, if the WDM channel uses some of the new proposed Data over Light protocols (SONET lite), such as synchronous 10xGigabit Ethernet, then a simple data bridge device residing at the customer premises (such as a Gigabit Ethernet switch) can be used to deliver tributary services at an add-drop point.
A Look Ahead
While an optical internet may become the dominant technology for transporting internet traffic, a spectrum of customer requirements will need to be met. Doing this while also meeting the demand for all-IP networks will become much more feasible with the advent of even higher density WDM systems, which can simultaneously support a multitude of transport service delivery mechanisms, from traditional SONET/SDH services to the new optical ATM and IP architectures.
In the next few years, Internet bandwidth will dramatically increase as these very high-density WDM systems are deployed. Perhaps the current amazing growth rate of doubling capacity every six months will even be surpassed. Such growth rates will have a profound impact on the future architecture of both networks and information systems, as the capacity of the network, not the computer, becomes the ultimate driver of system designs and advanced applications.
===========================================================
Andrew Bjerring is president and CEO of CANARIE, Inc. Contact him at andrew.bjerring@canarie.ca. Contact Bill St. Arnaud, CANARIE's senior director of advanced networks, at bill@canarie.ca. The Canadian Network for the Advancement of Research, Industry, and Education is a not-for-profit, industry-led consortium, tasked to develop next-generation, advanced networks and accelerate the evolution of the knowledge economy in Canada. |
