Fast NICs

By J.B. MilesSpecial to GCNThe fast lanes of network connections are getting faster'and just in time.The volume of traffic on most government LANs increases exponentially in response to the high-bandwidth requirements of scientific modeling, desktop publishing, data warehousing, video, and Internet and intranet applications. It's no secret that small, 10-Mbps Ethernet LANs cannot keep up with the demands placed on them.One cost-effective alternative has been to upgrade to Fast Ethernet networks using 100-Mbps switches, hubs and routers paired with dual-speed 10/100-Mbps network interface cards.Almost everybody wins with Fast Ethernet. Making the transition from standard 10-Mbps Ethernet is relatively easy because Fast Ethernet uses the same 802.3 Media Access Control layer interface, as well as the same packet lengths and error-control and management techniques. Fast Ethernet components are cheap to buy and the technology makes use of unshielded twisted-pair copper wiring already in place.But no matter how popular it is, Fast Ethernet isn't the only high-speed game in town. Competing systems, such as those using 155-Mbps and 622-Mbps asynchronous transfer mode, 100-Mbps Fiber Distributed Data Interface/Copper Distributed Data Interface and Gigabit Ethernet are also making their presence felt, particularly on network backbones.As for which technology is most likely to win the hearts and minds of government information systems managers, the bidding isn't over yet.If you already have an ATM, FDDI/CDDI or Gigabit Ethernet LAN or network backbone, the choice among the fast NICs listed in this guide is not difficult. Pick one that uses the technology you use, and then look for the feature sets that most closely fit your requirements.No matter what protocols they use, all NICs work pretty much alike. They provide a connection between a network component, such as a server, and the network itself. Without a fast NIC, your server cannot use the same Fast Ethernet, ATM, FDDI/CDDI or Gigabit Ethernet protocols used by the other equipment on the network.All NICs come with on-board processors, on-board RAM of various sizes, ports for connecting to the network and the host server, and usually management software. In addition to the type of protocol your system uses, your choice of NIC should be governed by other criteria: In Intel servers, 32-bit PCI buses are the dominant bus types, although 64-bit PCI buses are becoming more common. SBus components are used in many of Sun Microsystems Inc.'s popular Sparc servers and the NuBus is used in many Apple Macintoshes. Some of the NICs listed here are compatible with the 64-bit VME bus and, although a few still offer EISA and ISA bus compatibility, many manufacturers are dropping these in favor of more popular PCI models. The speed of a NIC is determined by the number of bits it can handle; 64-bit NICs process data faster than 32-bit NICs. Because most NICs are designed for the 32-bit PCI bus, they are prominent in the marketplace. Single Attachment Station NICs connect to a server via a single port. Dual Attachment Station units can connect to two servers via a single port, providing a degree of fault-tolerance in case an attached server goes down. Expect to pay about double for a DAS NIC. With a full-duplex option, a NIC can provide up to twice its advertised speed as long as it uses a dedicated connection to a network switch. ATM NICs are compatible with multimode fiber and Category 5 unshielded twisted pair copper cable, depending on the models selected. Gigabit Ethernet NICs do likewise, as do FDDI NICs designed to fit the new CDDI standard for FDDI products. NICs that connect with UTP-5 copper cable almost universally use a standard RJ-45 connector. Those connecting with single-mode fiber cable use ST connectors and those connecting with multimode fiber use SC connectors. The Media Independent Interface allows connections between fiber-optic and copper UTP cable. Most NIC manufacturers provide proprietary management software that usually is based on Simple Network Management Protocols or Desktop Management Interface protocols. With it, users can gain access to the NIC and other network components via an SNMP agent, Internet browser or Microsoft Windows console. Most NICs also come with drivers for a broad range of popular operating systems such as Microsoft Windows NT, Novell NetWare and various flavors of Unix. Introduced with Fast Ethernet NICs, bus mastering helps offload the work from the host CPU by passing any network requests directly to the NIC. The more memory a NIC has, the better and faster it can process information passing between the server and the network. NICs with 1M or 2M of RAM and fast processors are priced higher than units with similar specifications but less memory and slower processors. This used to be the province only of Fast Ethernet NICs, which can automatically distinguish between 10-Mbps and 100-Mbps data signals and set themselves accordingly. The newest Gigabit Ethernet cards distinguish between 10-Mbps, 100-Mbps and 1-Gbps data flows and work at the optimum level without user intervention. ATM and FDDI/CDDI cards lack this capability. Any NIC with Windows Plug and Play is automatically set up and configured by the Windows operating system.If you haven't decided on Fast Ethernet, ATM, FDDI/CDDI or Gigabit Ethernet, you have some difficult choices to make before selecting the right NICs for your servers.Simply stated, 100-Mbps Fast Ethernet can be the fastest, most cost-effective and easiest upgrade path from standard 10-Mbps Ethernet because the same network infrastructure can be used. Prices for Fast Ethernet components, including NICs, are dropping rapidly and account for the largest growth segment of the network market. But in some cases, choosing one of the following options could be as good or better: Not long ago, ATM was the darling of high-speed networking gurus. It was reliable, guaranteed quality of service and handled high-bandwidth video and voice traffic at speeds up to 155 Mbps and 622 Mbps.Proponents called it the glue for interconnecting heterogeneous networks into a larger network, and it was being widely touted as the most viable available technology for enterprise backbone networks.ATM is a switched architecture that creates virtual circuits for all attached network devices. These devices, including high-bandwidth servers, don't have to compete for available network bandwidth as they do under the various Ethernet schemes. And because it employs uniform, fixed-length 53-byte data cells, ATM is ideally adapted for use with any type of data flow, including voice and video.The ATM Forum recently firmed up some new standards, including a security specification, and did more work on voice, data and fax ATM transmissions.ATM scales well and is compatible with most other types of networks. And many of the compatibility problems among manufacturers' ATM products have been solved by evolving ATM standards.Critics, however, say ATM will never bring full 155-Mbps speeds to desktop PCs'25-Mbps ATM NICs are available for PCs and small workstations. Some say that ATM, once seen as the future of networking, will be relegated to a subsidiary role as a transport mechanism for IP traffic.In any event, users who are familiar with 10-Mbps Ethernet are comfortable with a 100-Mbps Fast Ethernet upgrade and are ogling Gigabit Ethernet for superfast data pathways. These users seem uninterested in making a changeover to unfamiliar ATM territory. FDDI, at 100 Mbps, was designed from the outset to be fast and reliable. Because the original specifications called for the exclusive use of fiber-optic media, which is difficult to tap into, a token-passing technique similar to token-ring was developed for transmitting data, but timing devices were added to give the system greater speed.FDDI traffic is carried over two rings, one primary and one secondary. The secondary ring is inactive, going into play only when a fault occurs in the primary ring. Devices or LANs can be attached to one or both rings. When both are used it creates a fault-tolerant system that can heal itself in case the primary ring, or cable, fails.CDDI results from adopting new standards for use of UTP-5 copper cable. Proponents of Gigabit Ethernet say it provides 1 Gbps of bandwidth for campus networks with the simplicity of Ethernet for a lower cost than competing high-bandwidth network technologies.Because Gigabit Ethernet uses the same protocols, frame formats and frame sizes as Ethernet and Fast Ethernet, it follows that existing networks can be extended to gigabit speeds at reasonable initial costs. Support staff and users don't have to be retrained. The result, according to Gigabit Ethernet backers, is a low cost of ownership for users.Recent marketing data supports the blue-sky predictions for the future of this technology. The In-Stat Group of Scottsdale, Ariz., predicts that the market for Gigabit Ethernet equipment will grow from $270 million in 1998 to $1.4 billion by 2001.Other indicators are that Gigabit Ethernet products outshipped ATM LAN products by a slight margin during the second quarter of this year and that the margin will widen in the future. Industry observers also note that the per-port price of Gigabit Ethernet installations is already less expensive than most ATM installations and will continue to drop.The recent passage of standards for copper-based Gigabit Ethernet using standard UTP-5 cables also will make it less expensive and easier to use for organizations that do not want to install fiber.Only a handful of Gigabit Ethernet NICs are being produced now, but expect more to come to market quickly, especially for those using UTP-5 copper cables. XXXSPLITXXX-XXXSPLITXXX-