re: Terabit [and Petabit?] Routing
From the Sept. 2000 Fiber Exchange articles:
"Terabit-scale Traffic Cops"
Is your network equipped to handle billions of packets per second?
By Kathleen Richards
fiber-exchange.com
-------begin story:
If you've ever tred to reach Boston via I-95 south during rush hour, or access popular Websites during prime-time Internet-usage hours, the nightmare of stop-and-go traffic crawling toward its destination is an all-too-familiar experience.
Picture this same scenario when Internet access becomes as ubiquitous as the telephone. Or when viewers at home can interactively play along online to win television game shows such as "Who Wants to be a Millionaire." Even without everyone online at the same time, it's no secret that carriers and service providers will need to manage high volumes of data traffic more efficiently and cost-effectively over the course of the next few years to compete in the open market.
Data is already the dominant traffic on the public-switched telephone network. It is growing 125% per year, while voice is increasing at less than 10%. At this rate, voice traffic will be less than 1% of total traffic by 2007, according to forecasts by market researcher Technology Futures Inc. (Austin, TX).
These traffic dynamics are bringing about a major change in the technology and the equipment that carriers use to handle the network transport of voice and data. Unlike voice, which requires a dedicated circuit or connection, data is "bursty" traffic that can be sent piecemeal over different routes to the same destination, freeing up unused bandwidth to other traffic on the same channels. Based on the price/performance efficiencies that ATM/IP cell- and packet-based technologies promise, many incumbent carriers are transitioning the technology in their networks from time-division multiplexing (TDM) or circuit-based switching to ATM cell-based and Internet Protocol (IP) packet-based switching. Today, ATM, which offers some quality-of-service (QoS) guarantees, is used largely in the network core while IP is a common protocol at the edge of the network.
IP routers already drive market
According to market-research firm RHK Inc. (South San Francisco), the multigigabit- and terabit-scale device market, which comprises both switches and routers, will almost double this year, reaching $2 billion by 2001 from $1 billion in 1999. "Most of that growth is clearly attributed to IP-based routers," says Raj Mehta, senior analyst of core switching and routing at RHK.
Computer-networking powerhouse Cisco Systems had the highest market share in 1999, followed by new IP-based router entrant Juniper Networks (Mountain View, CA). Juniper had its first commercial IP-based backbone router, the M40 (40-Gbit/sec throughput), available and in network trials in 1998. Today, Juniper is selling its product line, which now also includes the M20 (20 Gbits/sec) and the M160 (160 Gbits/sec) to carriers and service providers worldwide. In the next year, Mehta expects to see revenue from several more new entrants such as Avici Systems, Charlotte's Web, and Ironbridge Networks.
There is certainly opportunity in the market. By the end of 2003, the multigigabit- to terabit-scale device market is forecast at "well north of $12 billion," according to the RHK analyst.
Switch-routers
RHK is already combining Layer 2 and Layer 3 switches and routers into the same forecasts based on several trends in the market. One trend is the hybridization of these devices as carriers start to migrate to cell- and packet-based switching to more efficiently transport data. As various routing protocols start to work together and Multiprotocol Label Switching (MPLS) gains widespread acceptance among carriers and service providers, it is going to be more difficult to distinguish between a circuit-based or TDM-based switch and the packet-based router.
"MPLS is part of a simpler way of doing switching, because you are really looking at labels rather than looking at the whole IP destination address," says Claudio Mazzuca, product manager at petabit-scale router developer Hyperchip (Montreal).
"When you have MPLS, you can switch cells and you can switch packets," explains RHK's Mehta. "It is really futile trying to distinguish between a switch or a router; that is not the case today because there is a strong distinction between the two and what they do. However, within a year or two time frame, I think that the line will become blurred."
By and large, four factors-performance, port density, redundancy, and scalability-will differentiate the existing devices and next-generation switch-routers, according to Mehta.
In its simplest form, performance is how fast the router can look up and forward IP packets within a given system. "Juniper was the first to set the boundary high on that metric," says Mehta. "We expect someone like Avici to come in and push that performance boundary."
Avici Systems Inc. (Billerica, MA) has its terabit switch-router in network trials with AT&T, Deutsche Telekom, Enron Communications, and Williams. Both Enron and Williams have agreed to purchase the product.
Port density is another area of major concern to carriers. Space is already at a premium in carriers' collocation facilities, where most of the equipment lies, such as the central office or points-of-presence (PoPs). Thus, the goal is to squeeze a higher number of specific interfaces into a standard 7-ft rack or frame.
"If the vendor can collapse the two frames and put it within one rack, independent of these other three factors, then it is clearly a strong value-add to a carrier," says Mehta.
Redundancy, or availability, is the ability of the router to reroute traffic in case of loss and to guarantee a level of service comparable to what users have come to expect from telephone equipment today. "When was the last time you picked up your phone and didn't have a dial tone?" asks Mehta. "It's all relative. This metric is really about the amount of service uptime that you can expect a router to have, and in all fairness, it is almost three to four magnitudes off from voice equipment. The reliability of the router world is a long ways from voice equipment-it has to do with the fundamental architecture of the way IP works."
Nevertheless, many vendors are attempting to address service issues. Avici Systems, Hyperchip, and others are highlighting QoS features in their router products.
"A lot of the IP services carriers are trying to roll out need a certain quality of service such as video on demand and any other high-bandwidth and time-sensitive IP application," says Hyperchip's Mazzuca. "You have to have a pretty strict QoS mechanism throughout your network with low latency, so you want to make sure that feature is in the equipment that you roll out.
"What carriers are looking for in some ways is ATM-type quality of service, where they can just basically say, okay, we're building an IP network," he adds.
The final basic differentiator between products is the capability of the switch-router to grow or scale along with the network. How scalable a system needs to be is open to some debate. Today, the average router shipments are on the order of 100 Gbits/sec with some 160-Gbit/sec and 180-Gbit/sec devices, according to RHK research.
Carriers will not need terabit capacity in a single device for at least another few years. "I don't think anyone can disagree there; if you look at the average deployment today, it is going to be an order of magnitude off. A terabit-scale device is something that has the capacity to scale to a terabit, so even if it is a terabit-scale device, it won't ship with terabit capacity for a while in one single product," says Mehta.
Interaction with the optical layer
Hyperchip is one company already banking on the need for petabit-scale routing. That need will rest in part on the adoption of optical switching at the core. "Lambda switching will make the core more efficient in terms of bandwidth and restoration, because it is able to handle more bandwidth than the traditional ring architecture that is used," observes Hyperchip's Mazzuca. "So when carriers move to this mesh network, you will lose all the efficiencies that you gain if you try to scale your existing core routers to meet the bandwidth demand. Companies like Avici and some of the other terabit startups are coming in to alleviate some of the scalability problems for today and let's say the next couple of years, but then once you get over that tens of terabits needed at the PoP, you are really going to need another solution. With some of these crossconnects, you're getting huge port counts-some companies are saying 1,000-by-1,000 ports. What you need is the router to be to be able to feed the specific packets at the right wavelength in order for it to be efficient."
"We have terabit equipment that is scalable to petabit levels, so we've developed a switch fabric based on a massive parallel architecture similar to what used to be used in super computing, and applied that into the router space. So really, what we've developed is a switch architecture that can scale to over 65,000 ports, and these ports can be either OC-48, OC-192, or once the OC-768 standard becomes a bit more defined, OC-768 ports, as well as Gigabit Ethernet and 10-Gigabit Ethernet," explains Mazzuca. Hyperchip's router will be in alpha trials in the third quarter and beta in the fourth quarter of the this year.
RHK's Mehta believes that petabit-scale routing requirements in a single product are still quite a ways off. "I think of it as a gradual upgrade. I think the capacity requirements for terabit are still a few years off, and a petabit is a thousand terabits, so you do the math," he says.
Meanwhile, there is a lot of activity and development toward IP over the optical layer. That is evidenced in part by the initiatives of several standards bodies such as the Optical Domain Service Interconnect (ODSI) consortium, a group of more than 50 carriers and vendors, and the Optical Internetworking Forum (OIF), an industry group that is working to further standards. The standards in progress are being designed to ensure interoperability and intelligent communications between the IP routing and the optical-switching domains. That will allow carriers to provide dynamic bandwidth provisioning within minutes, for example, by signaling to the lower layers for more bandwidth.
end story-------
FAC |
