Physical Layer

Definition

The physical layer can be divided into the following main parts:

  • Network infrastructure

This includes cabling as well as active components such as hubs, switches, routers (these are not just pure hardware but have built-in intelligence).

  • Network Interface Card (NIC)

Every server or client machine in the network has at least one of these cards through which all network traffic must pass.

Error tolerance from the high availability viewpoint depends mainly on the network technologies used. In the past, fiber distributed data interfaces (FDDIs) were used to improve availability in local area networks (LANs). Recently ethernet technologies with redundant switches have become the favoured approach.

For more information, see Server Network.

Structure

Alternative Network Paths

This method relies on alternative network paths that can be used when the primary connection path fails. Network components such as switches and routers re-route network traffic. However, switching to alternative paths can take some time and is not always transparent to higher levels of the network. TCP/IP sessions often fail and have to be re-established. Therefore, a network design with switchover times shorter than the TCP connection timeout is desirable, so that existing connections are not affected when a network component fails.

We recommend using redundant network design. Although a certain switching time is required, you can avoid administrative effort to switch to an alternative network path and you can also avoid reconnects at the application level.

Network Interface Card (NIC) and Switchover Software

Even if the rest of the network is protected against failure using FDDI technology or redundancy in the network topology, the network interface card (NIC) remains a single point of failure in the physical part of the network connection. NIC failure causes loss of the network connected for an attached server. The NIC can in fact be considered as part of the hardware of the individual computer that it serves. Furthermore, redundant network cards are necessary to exploit the features of redundant switches.

If the NIC fails, one solution is simply to switch to a backup machine. Another solution is to install a second (standby) NIC in the same machine. Switching to a second NIC in the event of failure is faster than switching to a backup machine.

The handling of multiple NICs on a single computer is not easy. Special software products can be used to perform the takeover, by which the standby NIC takes over the address of the primary NIC in order for connectivity to continue uninterrupted.

One solution is to use proprietary switchover software. In this scenario the standby NIC is connected to the same physical subnet as the primary one. The switchover software constantly monitors the availability of the components. If a primary NIC fails, the standby NIC is automatically activated with the address of the primary NIC, and the connection resumes. The switchover often occurs so rapidly that applications with TCP/IP connections across the damaged NIC are unaffected. The performance and characteristics of such a solution depend on the switchover software product used.

There are also proprietary solutions that enable several network NICs to be addressed using the same IP address. You can use these to bundle the connection capacity of the NICs and also to reduce downtime.

Finally, you can reproduce the functionality of switchover software by writing your own scripts. You can use this to switch the address of a failed NIC to the remaining NIC.