Upgrading and Repairing Networks

Load Balancing and Dual-Redundant Network Controllers

For important servers, a single network card that is used to interface with the network is a single point of failure. If the card goes bad, the costs involved can be quite high if a large number of users access the server. For many years computer manufacturers have come up with solutions to minimize downtime for important systems. The two most basic techniques that come to mind are clustering and redundancy. Clustering technology allows two or more computers to operate in a manner such that one or more computers in the cluster can service user requests at the same time, as in HP's OpenVMS clusters. Another clustering technique, which is used in Windows 2000 Server and Windows Server 2003 clusters, uses a fail-over technique. One or more computers in the cluster are on standby and ready to take over if another member of the cluster fails.

This second technique is also an example of redundancy. For many years it has been possible to buy disk controllers that operate in a redundant fashion. Redundant node clusters, disk controllers, multiple CPUs, and other mechanisms have made the task of providing maximum uptime for computers an easy task. However, the network card remains a single point of failure.

That problem has been remedied. You can now purchase network adapter cards that work in a dual-redundant manner. That is, one card services network requests while the other waits in standby mode, checking periodically to be sure the other adapter is functioning properly. There are even versions of dual-redundant network adapters that both operate at the same time, providing a load-balancing capability as well as a fail-over mechanism.

For example, one manufacturer produces NICs that use a technique in which each NIC has its own MAC address, yet all cards that the administrator places in a "group" share the same IP address. Each group can contain from two to eight network cards, which should provide enough redundancy for almost any server. The added benefit of load-balancing to maximize network traffic throughput makes this an ideal solution for a high-end enterprise server that is used by a large client base.

Different methods can be used when load-balancing is implemented with network adapter cards. For example, one scheme is to use a round-robin technique for each packet sent out on the network. Another is to use the client's MAC address and assign traffic for those packets to a specific controller. This technique might not be a good solution if most of the traffic is coming from a router because all the packets from the router share the router's MAC address.

Although dual network adapters are not a factor in the desktop environment, they can provide security at that vital link in the network between an important enterprise server and the network. The cost of redundant network adapters, when compared to the expensive server hardware, makes this a cheap solution to a big problem. Because this technology has been around for only a few years, I suggest you check out several vendors, inquiring about the methods used for load-balancing and redundancy. Take into consideration also the management software, if any, that you will need to use to configure or troubleshoot the adapters.