Cheap switches will change the LANscape

W ith Bay Networks and 3Com locked in a deadly battle to decide which company makes the cheapest LAN switch, users face the happy prospect of getting LAN switching for less than a shared-media network interface card would have cost five years ago. At ComNet last month, 3Com announced a new version of its SuperStack II Gigabit Ethernet switch that sells for half the per-port price of the previous version. Meanwhile, Bay promised its BayStack 250 hub at only $75 per port. How low, as they say, can you go?

But will this cheap switching speed help or hurt the network overall? And does fast edge switching change some of our almost-sacred assumptions about LAN planning?

Let's address the second question first. For almost a decade, vendors have been telling us that the old 80-20 rule, which held that most LAN traffic was confined within the workgroup, was changing. We were supposed to be entering a new age in which LAN users sucked data from every nook and cranny of the business, destroying traditional traffic patterns.

Well gang, apparently the people buying workgroup switches didn't read about that trend because they're actually loading up network capacity and high-bandwidth applications more at the workgroup level than anywhere else. If the 80-20 rule is breaking down, it's possible that it's becoming more like 90-10 in favor of localized traffic.

A shift in LAN capacity reserve toward the workgroup would create a LAN with a sharp technical and political bias toward the edge user. Departments within organizations can buy cheap switches as readily as cheap PCs, thereby defeating the central LAN planning some companies still believe they have in place. That makes it harder to deploy LAN features that are designed to operate at the enterprise level, such as policy management of LAN quality of service (QoS), because all of the switches may not support the desired features.

Standards in areas such as QoS and management are notoriously difficult to develop at the pace the market requires. A shift in switch buying toward the network edge could push back adoption where standards aren't in place.

Increasing LAN capacity at the edge also creates potential problems with those applications that do operate across the enterprise. LAN applications don't have an effective strategy for dealing with congestion - data piles up until it gets discarded somewhere along the path. The loss of data signals applications to throttle back their flow control. That's as efficient as controlling your highway speed by accelerating until you run into the car in front of you, then stopping for a while.

In the past, congestion in shared-media edge LANs helped to keep applications from pumping megabytes of data into congested core LANs and WAN gateways. With edge congestion eliminated, traffic could pile up in second-tier switches or routers, exhausting buffers and creating performance problems.

Making a transition from shared-media LANs to LAN switching is a basic change in LAN architecture, not just a matter of gluing a few switches onto the network here and there. A switched LAN should be hierarchical, with each level providing increased switching capacity and faster LAN trunks.

Today's cheap workgroup LAN switches rarely are useful at the second tier of the switch hierarchy. A 10M bit/sec switched workgroup needs a 100M bit/sec switched second-tier LAN switch, and the third tier needs to provide gigabit-level capacity. Users often complain that low prices in workgroup switches aren't carried up to the higher tiers. Consequently, users are tempted to deploy the same 10M or 100M bit/sec switches at all the levels of the network, ignoring the converging traffic and increased congestion this creates.

But the real problem with architecting a switched LAN that will perform effectively for both workgroup and enterprise applications is procurement politics. It's easy to see who has to buy a workgroup switch - whoever runs the workgroup. Second- and third- tier switches don't have such convenient and direct associations with a management level, and so often the "LAN group" is expected to buy them. But the LAN group doesn't make or sell the company's product and often finds that getting the capital budget to enhance the network's core is a tough proposition.

In a recent survey by CIMI of almost 300 companies conducted by my organization, six out of 10 workgroup network managers said they could purchase local switching for their workers simply by signing for the equipment. Five out of every six central LAN managers in the same survey said they had no capital budget for switching at the second tier and higher. That's as logical as buying electrical appliances without installing outlets.

Line departments at the network's edge are gaining control of the network's purse strings and its evolution - at least for their local part of the network. This is because companies are seeking more direct business justification for technology purchases, and that requirement is easier to meet if you're a department that's really conducting business. As a result, the network staff or infrastructure technology groups are finding it harder to get money. If this trend continues, progress in the core may cease and enterprise application goals - including intranets - may be compromised.