In previous tips we have discussed various layers of the availability model. This tip looks at availability issues...
involved in a network.
No component demonstrates that availability is a chain better than networking. Other components are generally singular; there is a client, a keyboard, a disk array, a data center, a server -- or two or three. But those components are well-delimited and easy to define.
A network is a much more complex beast; it runs places where it cannot always be secured, across the equipment of multiple vendors and across the wiring of many different providers.
Did you know, for example, that if you purchase two parallel WANs from different vendors to run from one site to another, it is extremely likely that somewhere along the line they will run in the same switching station, possibly in the same conduit and maybe even in the same multiplexed cable? It is quite likely to be true, and there's almost no way to find this out. Different network providers often lease or buy cable from a single source. Even though you may not know they are, that single source can be a single point of failure for your entire network.
Networks run through unoccupied cubicles, and empty offices. Would you leave your wallet sitting on your desk when you leave your cubicle? Presumably not. Yet companies leave network connections unprotected all the time. If an unscrupulous person can gain access to an unprotected network port inside the firewall, then he can gain access to any of the data inside that firewall as it flows across the network.
One bad network component can shut down the network for everybody. Once, a very long time ago, your humble author plugged a telephone into the wrong port in his cube and brought down the entire office network. The network recovered just fine as soon as the phone was unplugged, but if I hadn't realized what I had done, or if I had had some malicious intent, I might have left the phone plugged in. Depending on the sophistication of the network infrastructure, it can be very difficult to determine what -- or more importantly, where -- the problem is.
It's very difficult to define exactly where the network is. The path from a PC to the server in the machine room may pass through a switch or LAN concentrator in a wiring closet, a router or two in the same building, another router entering the data center building and one or more machine-room high-speed external networking interface. Then there are the network cards on PCs, switches, routers and servers. There are dozens or hundreds of potential failure points. And the next question is who manages each of these components. Depending on the organization, each type of component could be managed by a totally different group and depending on local politics, these groups may not speak to each other.
As with so many other technologies, the key to network reliability is simplicity. Keep the design simple. Limit the number of components and you'll limit the number of potential failures. Limit the number of potential failures and you will limit the number of actual failures.
Copyright 2002, Evan Marcus
Evan L. Marcus is the data-availability maven for Veritas.