Networking

Choose your NIC wisely

Trying to figure out what type of NIC you should use for your network? Read Rodrigo Zamora's insightful take on the features and benefits of network interface cards.


Since networking has permeated almost every aspect of computing, support technicians need at least a fundamental understanding of network components and how to troubleshoot them. Among the most ubiquitous of these components are the Ethernet network interface cards (NICs). As a support technician, you will have to troubleshoot, upgrade, and perhaps even design an Ethernet network. You may also select and replace network cards, when needed. To do any of these tasks effectively, you should familiarize yourself with the features of Ethernet NICs.

Before I begin
For the purpose of this article, I will define a NIC as an Ethernet adapter card intended for installation in a workstation or server system bus slot; however, a NIC doesn’t have to be an adapter card. (There are NICs designed as external devices that connect to parallel or USB ports and also network interfaces integrated onto PC motherboards.) I will also assume you possess a fundamental knowledge of Ethernet networks and terminology. If you are not up to speed on the basics, there are many excellent introductory articles and tutorials available on the Internet, including some fine ones on TechRepublic. For a basic introduction to networking and Ethernet, see Lantronix’s Ethernet Tutorials.

Network card functions
In terms of the OSI Reference Model, a network card can be thought of as part of the Data Link layer (Layer 2) and of the Physical layer (Layer 1). The NIC’s functions include providing a physical address (MAC address) for a host on a local network, a method for encapsulation of data into frames, a means for encoding the frames into a bit format suitable for the local network, and the placement of the frames onto the physical medium. A NIC also uses a media access control (contention) protocol to deal with collisions that occur when more than one device attempts to communicate at the same time. Ethernet’s media access control protocol is Carrier Sense Multiple Access with Collision Detection (CSMA/CD).

Speed and connectors
The Ethernet variants that a NIC supports determine its theoretical maximum speed. There are several IEEE Ethernet variations that support different speeds and media. The most common choices for workstation NICs are 10Base-T and 100Base-TX. 10Base-T is a 10-Mbps, baseband network connected in a physical star topology using twisted pair cabling. 100Base-TX is a 100-Mbps network also connected in a physical star using twisted pair cabling. Even if you are installing a NIC for use in an older 10-Mbps network, you should buy one that supports 100Base-TX. There is usually no cost difference between the two, and the 100Base-TX provides an upgrade path if you decide to expand to a 100-Mbps network. Also, many dual-speed NICs support both of these network types through a single RJ-45 jack.

10Base2 (aka ThinNet) and 10Base5 (aka ThickNet) are 10-Mbps networks connected in a physical bus topology using coaxial cable. The difference between these two networks is the type of coaxial cable they use and the resulting maximum segment length—10Base2 uses a smaller diameter cable (RG-58). To connect a coaxial cable to a 10Base2 adapter card, you must first attach a T-connector to the adapter’s BNC connector. The coaxial cable attaches to either side of the T-connector. If the NIC resides on a host at the end of the bus topology, a terminator resistor must be attached to one side of the T-connector. Figure A shows some of the cables and accessories that connect a 10Base2 network.

Figure A
Shown clockwise from the left are two UTP cables, an RG-58 coaxial, a terminator cap, and a T-connector. (Photo courtesy LinkSys Group Inc.)


10Base5 NICs attach to a network by connecting a drop cable to its 15-pin AUI connector (DIX connector). The drop cable then attaches to a vampire tap transceiver, which taps a ThickNet cable. LANs using coaxial cable such as 10Base2 and 10Base5 are no longer recommended and thus, virtually obsolete. However, you may have to upgrade an older network that still uses these Ethernet types. In such cases, you might consider a combo-card that includes 10Base2, 10Base-T, and sometimes 10Base5 connectors. Figure B shows an ISA combo-card NIC that supports 10Base2 and 10Base-T.

Figure B
Notice this card offers both a BNC connector and an RJ-45 jack. (Photo courtesy LinkSys Group Inc.)


On NICs intended primarily for servers, you may encounter Ethernet variants like 100Base-FX (aka Fast Ethernet) and 1000Base-X (aka Gigabit Ethernet). 100Base-FX uses either single-mode or multi-mode fiber optic cable as its medium and, like 100Base-TX, it has a maximum bandwidth of 100 Mbps. Ethernet NICs using fiber cable will have at least two connectors (usually SC or ST connectors) so two fiber cables can be attached—one cable for transmitting and the other for receiving.

1000Base-X is a relatively new variation of Ethernet that is primarily fiber-based (although UTP cabling is also supported). It’s capable of a maximum bandwidth of 1000 Mbps and is compatible with 10-Mbps and 100-Mbps Ethernet networks; however, the hardware requirements of Gigabit Ethernet are quite substantial (read: expensive). Therefore, you are not very likely to see or use Gigabit Ethernet other than in networking backbones in mid-to-large enterprises. As costs go down and bandwidth requirements increase, however, it’s likely to become more popular, so you should familiarize yourself with it.

Gigabit Ethernet requires a larger frame size to reduce collisions that would occur because of its increased speed. In 10-Mbps and 100-Mbps Ethernet variants, the minimum frame size is 64 bytes, which is the minimum size required by the CSMA/CD protocol to ensure that collision detection occurs by the transmitting host and the most distant host on its segment. In other words, data must remain on the medium long enough for the most distant host on the segment to sense it and for the transmitting host to sense any data sent by the distance host that would result in a collision. This minimum frame size has been increased to 512 bytes for Gigabit Ethernet. If less data needs to be sent, the frame is padded, which can result in inefficient network use when many small frames are sent. Gigabit Ethernet attempts to alleviate this inefficiency by providing packet bursting, where the NIC pads only the first frame of a series of small frames.

Data bus type
The majority of NICs on the market are PCI-based adapters because PCI is the dominant system bus standard due to its presence, speed, and bus-mastering capability, which allows it to minimize CPU utilization. PCI adapters can come in 32- or 64-bit versions. The 64-bit PCI NICs are used in servers or fast workstations and fit into either a 64- or 32-bit PCI slot. Note, you should consider using a non-PCI adapter only if you will be installing a NIC on a legacy system that doesn’t have PCI slots or a system that has run out of PCI slots. Ethernet adapters can also be found for ISA, EISA, and MCA data bus types. While ISA adapters are still readily available, EISA and MCA NICs are becoming increasingly rare.

Drivers
When selecting a network card, make sure stable drivers are available for the OS with which the card will work. Without proper driver support, the NIC will be useless in your setup. Having the proper drivers is especially important when upgrading a workstation to a newer OS version (such as from Windows 98 to 2000) or when you want to use the card with Linux. While most network cards now come with decent Windows 2000 drivers, not many come packaged with Linux drivers. Thus, you will have to see if some are available directly from the manufacturer or if the card has been confirmed to work with available open source drivers.

Full duplex
Most new Ethernet NICs of the 10Base-T, 100Base-TX, 100Base-FX, and Gigabit varieties support full duplex operation. Standard Ethernet is half duplex. It was designed to allow only one host to transmit on a medium at a time to prevent collisions. However, when a switch is used to connect hosts (effectively breaking up collision domains), only a host and the switch can communicate in each segment. Furthermore, since twisted pair and fiber variants of Ethernet use separate wiring for sending and receiving data, no collisions can exist in the switched segments. This means that a host can be sending data in one channel at almost full bandwidth speed while receiving data on the other channel at almost full bandwidth speed. This full duplex operation has the effect of doubling the network bandwidth when a host communicates through a switch.

Auto-negotiation
The Auto-negotiation (N-Way) standard allows a NIC to automatically set itself to work at the fastest speed of the network to which it is connected. If the NIC is connected in a segment in which the device on the other end is also auto-negotiation capable, both devices will exchange information about their capabilities and both NICs will set themselves to the highest speed supported by both devices. If the device on the other end is not auto-negotiation capable, the NIC will attempt to recognize the signal sent by the other device and set itself accordingly. This is called parallel detection. An Ethernet NIC with auto-negotiation can set itself to work with 10Base-T or 100Base-TX at half or full duplex or with 100Base-T4 at half duplex. Auto-negotiation is not supported by 100Base-FX.

Wake OnLAN
Some computer power supply types, such as ATX, provide constant standby power to a system’s motherboard, regardless of whether the system is on or off. In most cases, this power can be used to power devices that turn on a system when a certain event or condition is detected. Such devices include NIC cards that support the Wake On LAN (WOL) feature. A system implementing WOL can be remotely powered by sending its NIC a wake-up packet. This can be very handy to remotely manage systems during nonbusiness hours. To use WOL, the NIC and the system’s motherboard must support it, the NIC must be connected to the motherboard’s WOL connector via a three-pin cable, and management software capable of sending wake-up packets must be used.

Boot ROM support
Many NICs come with a socket that allows installation of a boot ROM. A boot ROM allows a system to remotely boot up without the need for a hard drive or a local copy of an OS. This feature makes it possible to use diskless workstations with Remote Initial Program Load (RPL) to load operating system files from a server. A boot ROM can also provide preboot services with the Preboot Execution Environment (PXE). With PXE, you can remotely and automatically boot, configure systems, and provide other system management services. For example, you could use PXE to install OSs and application software on several networked machines at once, provide disk images to reconfigure machines with problems, or remotely scan for viruses without having a technician visit each machine. You can find more information on PXE here.

Intelligent Interrupt Management
NIC cards use several mechanisms for reducing CPU utilization, like Intelligent Interrupt Management. The more hardware interrupts the device issues, the more often a CPU has to deviate from its current task to service those interrupts. A NIC card can reduce these interruptions by coalescing interrupts during periods of heavy network activity. This involves collecting more data before an interrupt is issued.

VLAN support
A virtual LAN (VLAN) is a group of computers that may be on physically different network segments but are logically grouped to communicate as if they were on the same segment. VLANs are implemented by using switches that support VLAN technology. Each VLAN becomes its own broadcast domain, so broadcast messages that are intended for a particular VLAN are not seen by hosts of a different VLAN. For hosts in different VLANs to communicate with one another, their messages need to pass through a router. However, servers using network adapters with VLAN support can belong to several VLANs simultaneously, eliminating the need for routing to be performed on messages sent to and from hosts that are members of those VLANs. For more information on VLANs, see this technical brief.

IPSec encryption offloading
IPSec is a Layer 3 encryption and authentication standard used to provide secure communications between end systems. It is particularly useful for securing data over a VPN, especially when it is used in conjunction with the Layer 2 Tunneling Protocol (L2TP). Some NICs can process the encryption of IPSec packets, thereby reducing the load on a system’s processor. This can be a useful feature when setting up a VPN server.

Adapter Fault Tolerance (AFT)
AFT allows a server to switch from one network card to another when a failure occurs to any component in a link. This prevents a failed card, cable, or connector from interrupting access to a server.

Load balancing
A network server using NICs capable of load balancing can eliminate bottlenecks by distributing network traffic evenly among its adapters. This also has the same effect as using AFT, because each adapter provides a separate link from the server to the network.

Hot swappable (PCI hot plug)
Some high-speed NICs intended for servers can be installed and removed without having to shut down the system. This feature, called hot swapping, helps minimize downtime by allowing faulty devices to be replaced immediately.

Jumbo frame support
This is a nonstandard method of using larger frames (9,014 bytes) with Gigabit Ethernet to reduce CPU usage and provide increased throughput. By using larger frames, there are fewer frames to process. To use jumbo frames, however, both of the communicating computer’s NICs and any networking devices between them must support jumbo frames.

Conclusion
Whether you are looking for an inexpensive means of connectivity for a small network or a very expensive, fault-tolerant solution for an enterprise, the complexity and feature base of modern NICs is impressive. By understanding the common features of Ethernet NICs, you can learn a lot about Ethernet networking technologies. Nevertheless, this article offers only a glimpse of the giant that Ethernet networking and its related technologies has become.

Editor's Picks

Free Newsletters, In your Inbox