LINX to Cisco: "Good Riddance"
Keith Mitchell, executive chairman of LINX, says Cisco hasn't delivered the goods required, and has been uncooperative. "Cisco focuses on the bottom line only," says Mitchell. "They don't perceive us as a high-volume customer. We're not significant enough for them to bother with." At press time, Cisco had not responded to inquiries from Light Reading on the situation.
LINX's chief beef with Cisco is that its switches are still based on an old school design. "Cisco is still producing blocking architectures for its switches," he says. "Our traffic doubles every hundred days or so. The only way to guarantee performance in this environment is through non-blocking switches." In switch architectures, blocking is a condition in which the overall switch fabric may not support fully loaded conditions, forcing some traffic to wait until a path reopens. Non-blocking architectures are typically constructed so that no delay will result, even if traffic is running full duplex over all connections.
LINX started its network in 1994 with several Cisco Catalyst 1200 switches, supporting 10-Mbit/s switched Ethernet links in its five-member London network. By 1996, the network was ready for an upgrade. A Cisco Catalyst 5000 was installed to support multiple 100-Mbit/s Ethernet links. By the summer of 1998, the winds of change blew again. For one thing, growing traffic levels threatened network performance. "Broadcast traffic was having a noticeable impact on the resources of the Catalyst 5000," says Mitchell. Further, the LINX membership--now comprising close to 50 ISPs from eight countries--voiced a clear demand for higher volumes of gigabit Ethernet.
LINX turned to Cisco for help, since the vendor's installed switches couldn't handle gigabit Ethernet at the volume the exchange required. "They told us to buy new Catalyst 5500s," Mitchell recalls. But the LINX team was worried that the Cisco products' bus-based architecture would act as a bottleneck at the higher, gigabit rates. This drawback threatened to force LINX to buy more switches than planned in order to accommodate the bandwidth demands of its members at acceptable performance levels. When confronted, Cisco personnel decried the importance of having a non-blocking switch LINX says. And they refused to share plans for future switch products with LINX.
Mitchell turned to other players for help. Extreme Networks answered the call by delivering four non-blocking, shared-memory Summit48 switches, followed by a BlackDiamond 6800. "Extreme was very receptive to us, and they were willing to listen to our ideas and let us know in advance what their plans were," Mitchell asserts.
The cooperative relationship with Extreme only highlighted the problems with Cisco, he says. The LINX staff decided to start replacing Cisco equipment with Extreme gear. Within six months, the Cisco 1200s were wheeled out of the co-location facilities, leaving the Summit48's in charge. Only the Catalyst 5000 remained, on limited tenure. By the start of the year 2000, the BlackDiamond replaced that switch. LINX also is testing gear from Foundry Networks Inc. http://www.foundrynetworks.com to provide redundancy for the Extreme switches.
Mitchell says Cisco didn't try to pursuade LINX from pulling its gear. "We told them. Maybe they'll say we haven't spoken to the right people. But it's their responsibility to put the right people in touch with us."
Meanwhile, LINX is on the move. The exchange now supports 100 member ISPs in three co-location facilities. By the end of April 2000, LINX plans to close contracts for three more co-location facilities. Traffic volume is up. And as far as Keith Mitchell is concerned, the relationship with Extreme Networks and Foundry is on the right track. "Extreme knows the value of what we are doing here," he says. "And we think their product direction supports our future goals."
-- by Mary Jander, senior editor, Light Reading http://www.lightreading.com