Ethernet overload worries IS

Workgroups buying cheap bandwidth could cause a strain on the network backbone

By Stephen Lawton

IS managers traditionally have been irritated by LAN groups that purchase desktop systems according to their own inclinations. Today, as networks support their organizations' mission-critical applications, this penchant becomes more than an irritation.

Some trends encourage this autonomy. For example, fast Ethernet (100Mbps), which users welcome for faster printing and file-sharing within their subnet, is now priced at what used to be considered 10Mbps levels.

Retail prices for 10/100Mbps NICs have fallen to as little as $30. Many PCs now ship with 10/100Mbps capability built onto their motherboards; buyers pay negligibly more for this feature, which eliminates the add-in NIC.

Additionally, switches and hubs are falling in price, giving department managers with purchasing authority the option of buying higher-performance network devices without necessarily taking into account the devices' impact on the corporate network. Although switches generally are considered wiring-closet devices and therefore within the purchasing purview of IS, some manufacturers are building small desktop versions that let local workgroups improve their file and print services.

Many department managers are opting for 100Mbps NICs rather than switched 10Mbps--either on the motherboard or as an add-in--because of the 80/20 rule about LAN vs. WAN usage. That rule still applies, says Dwane Shirakura, senior analyst at The Dell'Oro Group, a consultancy in Portola Valley, Calif.

These managers realize that most file and print services still are done on local subnets, and therefore their users will get better local performance if they go to fast Ethernet.

However, not all observers respect the 80/20 rule. In companies that rely heavily on intranet-based applications and groupware, the amount of work done on local subnets might be considerably less than 80 percent, says Gordon Stitt, president and CEO of Extreme Networks Inc. in Cupertino, Calif. In those cases, he says, it is common to find higher-performance switches in the workgroup's wiring closet for gaining access to the corporate backbone.

In small and medium-sized companies, department managers may have the purchasing authority for point products, even if they can also serve as infrastructure devices, Stitt says. There the gigabit ports of high-speed switches such as Extreme's might not be used, because the switches might have been purchased for their 10/100Mbps capabilities.

Assault on the backbone
With 100Mbps capability readily available at the desktop, bandwidth management and shaping are becoming more of an issue, Shirakura says. Some corporate IS managers are concerned that if they let local managers dictate the uplink speed of the local switch, the network backbone's performance may be compromised. If too many subnets start pushing data at 100Mbps, the 100Mbps backbone will become the bottleneck, forcing either major changes in the workgroup or an upgrade of the corporate backbone.

Some IS managers are making efforts to control the proliferation of 100Mbps devices by setting corporatewide restrictions on purchases by managers outside of IS, according to Greg Lang, business unit manager for network infrastructure at Intel Corp. in Santa Clara, Calif. In some cases, these restrictions are put forth as a list of approved products for departmental use; if a department manager buys a product not on the approved list, IS might not support it.

Some observers suggest that IS managers draw a line at the access devices--the wiring-closet switches and hubs closest to the workgroups. Although a workgroup manager might want local users to have 100Mbps access within their subnet, IS should dictate the bandwidth from that access point out to the backbone. That way, local traffic can get local benefits without negatively affecting the backbone.

The Dell'Oro Group's Shirakura advises IS managers to decide on a unified corporate approach to infrastructure devices. "Networks [today] are predominantly Layer 2-based flat networks," he says. If department managers are allowed to usurp the IS manager's ability to control bandwidth on the network, the IS department "will be forced to subnet [the network] or use Layer 3 routing switches," he says.

This could force companies into infrastructure upgrades before they are ready to implement them or willing to pay for them. It's also a major reason that gigabit Ethernet vendors hope workgroups will migrate to switched 100Mbps from switched or shared 10Mbps. In their view, if a 100Mbps Ethernet backbone begins to be overloaded because more users are grabbing greater amounts of bandwidth or because of the proliferation of 100Mbps uplinks from workgroups, the next step is to build a bigger backbone with gigabit Ethernet.

However, vendors often neglect to mention that adding gigabit Ethernet also requires new server NICs and switches, as well as a fiber-based cabling plant. At present, gigabit Ethernet is specified only for fiber outside the computer room; a 1000Base-TX specification isn't likely to be ready this year.

Even if IS dictates what NICs will be authorized, those decrees could go unheeded if the department manager has PC purchasing authority. Most PC systems with Ethernet preinstalled that are sold through the reseller and retail channels today include fast Ethernet. The corporate user has the option of using the onboard port or disabling the onboard device and installing an add-on NIC.

Onboard or add-in
Buying an Ethernet adapter on the motherboard has benefits and drawbacks. On the plus side, purchasing systems with Ethernet on the motherboard can eliminate the need to buy a separate NIC, nominally reducing the overall system cost, says Diane Meyer, an analyst at In-Stat Inc., a market-research company in Scottsdale, Ariz.

Additionally, Meyer says, when the NIC is integrated into the system, users should have fewer driver and installation issues, which will result in greater reliability.

Smaller companies without a full-time IS manager might prefer this approach, because the person buying the systems might not know the various NIC vendors or their respective strengths and weaknesses. Both 3Com Corp. and Intel, for example, supply silicon to systems manufacturers that include embedded NICs, Meyer says. They have strong brand-name recognition, and some systems vendors promote the networking brand almost as strongly as the CPU brand.

On the negative side, Meyer says, "the main issue is, what happens if the chip fails?" Replacing a standalone NIC requires opening the PC, disconnecting the Ethernet cable, and swapping the card, but the embedded approach requires replacing the motherboard--and that requires more time and technical expertise.

The number of systems with embedded fast Ethernet on the motherboard is growing, according to a recent In-Stat report that covers the first nine months of 1997. It showed that fast Ethernet on the motherboard grew by 80 percent: from 1.76 million units in the first quarter of 1997 to 3.17 million units in the third quarter. Meyer expects the figure to reach 10.28 million in 1997, grow to 18.1 million this year, and surpass 25 million by 2000.

Among the systems manufacturers currently offering fast Ethernet on the motherboard are Compaq Computer Corp., Dell Computer Corp., Hewlett-Packard Co., and IBM (Apple Computer Inc. was the first PC vendor to offer LAN on the motherboard, but its fast-Ethernet offering requires an optional NIC).

In a contrarian move, Compaq recently changed its product mix, essentially dropping embedded NICs because most IS managers opted to buy a separate NIC as well, says Michael Takemura, Houston-based product-marketing manager for Compaq's Deskpro products in North America.

Compaq accomplished this change by relying less on preconfigured systems and by allowing resellers who work with small businesses more say about the way Compaq configures the systems it sells, Takemura says. Compaq also plans to offer a build-to-order program, similar to the strategy used by Dell and Gateway 2000 Inc., in which retail customers choose Ethernet on the motherboard or as a NIC.

In the embryonic market for thin clients, having the Ethernet adapter on the motherboard is a requirement because of the lack of space for an add-in NIC, says Justin Smith, an analyst at International Data Corp., a market-research company in Framingham, Mass.

For traditional desktop systems, Smith says, users have more flexibility with add-in cards, such as being able to install multiple NICs for redundancy or change NICs as new features are added.

Are PC manufacturers in effect pushing workgroups to 100Mbps? In many cases, the answer is yes. Today's NICs, whether embedded on motherboards or added on cards, are moving to 100Mbps because the cost of supplying it is small, says Compaq's Takemura. Vendors say there is no sinister plan to boost desktop bandwidth; they're simply meeting customer demand. Whether this demand is always good for existing networks is a matter for IS managers to debate.


Copyright 2001- 2002
All trademarks are the property of their respective companies.