November 6, 2000
Marconi Communications PLC (Nasdaq/London: MONI) today announced a high capacity switch router called the BXR-48000 (see Marconi Announces Big Switch/Router).
The announcement could set the cat among the pigeons on a couple of counts. First, the switch is very high capacity. “It’ll be the fastest thing on earth,” according to Geoff Bennett, director of technology in the office of Marconi’s CTO.
Second, in demonstrating just how fast the BXR-48000 really is, Bennett points out that the capacity claims of competing vendors such as Avici Systems Inc. (Nasdaq: AVCI; Frankfurt: BVC7), Cisco Systems Inc. (Nasdaq: CSCO), and Juniper Networks Inc. (Nasdaq: JNPR) need -- how shall we put it -- careful examination.
Still, let’s start by seeing what Marconi’s got to offer. The BXR-48000 is a single stage, shared memory, non-blocking, 480-Gbit/s switch router. Initially, Marconi will offer a 240-Gigabit version in a single rack. Later on, it’ll offer a way of putting two systems side by side and linking them to a common switching fabric to deliver 480 Gigabits.
”It’s a direct descendent of the Fore Systems switch line,” says Bennett. The B in the BXR stands for "broadband" (as opposed to the “A” prefix of Fore Systems’ ATM switches). It signifies that the switch can handle both packets and cells natively. In other words, if a packet-over-Sonet packet arrives at the BXR-48000, it’s handled as a packet all the way through. Likewise with ATM cells. This differs from most other monster switches and routers, which either segment packets or encapsulate them, according to Bennett. Segmentation and encapsulation both exact a performance penalty, he adds.
Marconi is aiming to demonstrate the BXR-48000 in the first quarter of 2001 and start customer trials in the third quarter. General availability is planned for the fourth quarter of 2001. “This is very much a pre-announcement,“ admits Bennett. “With a box that has a price tag of $1 million, the sales cycle is very long.” Service providers need time to design their networks around the BXR-48000, he adds.
In some respects, it’s unfair to compare Marconi’s promises with products that are already shipping from vendors like Avici and Juniper. But in making such comparisons, Bennett highlights a number of ways in which vendors commonly overstate capacity claims.
For a kickoff, Marconi claims that its switch router will handle a genuine payload of 480 Gbit/s. Router vendors, Bennett contends, typically calculate capacity by multiplying the number of ports by their capacity, ignoring two issues. First, each input port feeds traffic to an output port, so counting both of them overstates the capacity by a factor of two. In other words, Juniper’s M160 can carry a maximum of 80 Gbit/s -- 80 Gbit/s going in and 80 Gbit/s going out.
Second, there may be other bottlenecks that make this calculation irrelevant. The backplane of Cisco’s GSR 12000, for instance, can’t handle all ports working flat out at the same time, according to Bennett.
In addition, some switch vendors overstate backplane capacity by not giving the genuine payload. They give the aggregate traffic that the backplane can handle, which includes overheads such as headers used for internal housekeeping, says Bennett.
Bennett says vendors making switch routers from clusters of equipment under a common management system, like Avici and Pluris Inc., also overstate their capacity by simply aggregating the capacity of each member of the cluster. This ignores the amount of bandwidth consumed in internal communications. It’s a bit like saying that a car can do 240 mph because each wheel can do 60 mph. In any case, says Bennett, “Avici isn’t shipping the clustered version of the box yet.” It’s only shipped individual 80 Gbit/s units, he says.
For many carriers, whether or not switches can route traffic from input to outport port in a single stage is an even bigger issue than capacity, according to Bennett. If switches aren’t single stage, then the queuing that’s unavoidable at output ports on the first stage translates into queuing at the input ports of the next stage, and that’s a big no-no, says Bennett. If traffic is queuing to get into the next stage of the switch, then other traffic is blocked -- resulting in delays.
Bennett maintains that a lot of switch routers -- among them Avici, Juniper, and Pluris -- aren’t single stage and thus can’t scale without running into blocking problems.
This is a crucial issue for all types of carrier, according to Bennett.
Incumbents are in a position where they’re making most of their money from telephony and leased lines, and want to run these over infrastructure that can also support a drive to offer Internet services.
"New age" carriers, on the other hand, are in the opposite position. Right now, many of them are losing money on plain vanilla Internet services and want to start offering high value services by guaranteeing quality. However, they’ve got a problem, according to Bennett. Most of them have installed switches and routers that aren’t genuinely non-blocking so they can’t control delays, the key to guaranteeing quality. A lot of effort is going into developing technologies that aim to prioritize different types of traffic -- but the fundamental problem won’t go away.
Bennett says Marconi’s BXR-48000 solves this problem by enabling service providers to divide switching capacity so that some can be used for handling premium grade services requiring guaranteed bandwidth while the remainder can be used for cheap-and-cheerful best effort services. This also enables incumbents to consolidate services onto a common backbone, thereby saving money.
At press time, Avici, Cisco, Pluris, and Juniper had not replied to requests for comment on Bennett’s claims.
-- Peter Heywood, international editor, Light Reading http://www.lightreading.com
You May Also Like
5G Network Automation and AI at Global Megaevents: A Telco AI-at-scale case study with Ooredoo and EricssonOct 10, 2023
5G Transport & Networking Strategies Digital Symposium.Oct 26, 2023
Improve Service Efficiency in the Call Center and Field with Slack AutomationOct 13, 2023
Open RAN Evolution Digital Symposium Day 1Jul 26, 2023