We focus primarily on software when we talk about scalability, but software runs on hardware, and the underlying machines can severely affect the scalability of the software that you implement. If you have to buy new hardware to run the latest operating systems or applications you need, it will add considerably to the cost of deploying the new software.
We all know that Moore's law renders every new machine outdated almost as soon as you buy it, but if you look at new hardware purchases with scalability in mind, you can make hardware last longer, and save money. Let's take a look at some scalability issues to consider the next time you buy new client computers, servers and network devices.
Scalable clients: the hard way and the easy way
Many companies will delay upgrading to Windows Vista because they're afraid their hardware won't handle the new OS. Although Vista will run on desktop systems that are available today at modest prices (minimum supported specs: 800 MHz processor, 512 MB RAM, 800x600 VGA GPU, 20 GB hard disk), many older client computers don't meet the criteria, especially the memory requirement. And these are just the minimums to run the basic Vista OS; in order to get all the "eye candy" features of the Aero Glass interface, machines will need a good graphics card — not the onboard video that many business desktops are using.
Even if you're willing to install new graphics cards in all the machines, it may not be that simple for older computers. Most of the Vista-capable video cards come with AGP or PCI Express interfaces. If all you have are regular old PCI slots, you may have problems finding compatible cards. If your users need Vista's features, you may be required to buy new machines or at the very least, spend some money on memory and video cards.
There is, however, an easier way to address the client scalability issue. Using the "thick thin client" model, your users can stick with their old, low powered PCs and still run a modern operating system. You do that by implementing a Windows Terminal Server; users' PCs become smart terminals that log onto the terminal server and do most tasks on the server. A number of improvements to Terminal Services are included in Longhorn Server.
If you choose not to use terminal services, probably the most important scalability issue to consider when you buy client systems is the ability to upgrade memory. You'll note from the specs for Vista that it will run on an 800 MHz machine, which was available in 2000. However, the typical desktop system in 2000 had considerably less than the 512 MB of RAM that's the minimum for Vista. With a few exceptions, new software needs more memory far more than it needs a faster processor, so be sure there are sufficient slots to add memory in the future.
Scalable servers: many ways to go
When buying traditional servers, the same caveat applies: make sure there's plenty of room for growth in the RAM department. With servers, it's also a good idea to allow for processor expandability, too, especially for those running databases or other services that require intensive number crunching. You can buy servers with motherboards that support multiple processors, but which only have a single processor installed, then add processors as they're needed.
Disk space is another issue for servers as your network grows and usage increases, but with today's external disks and external RAID arrays, as well as network storage solutions, this is less of a concern. Storage is not limited by the motherboard as memory and processors are.
For better scalability, you can look beyond traditional tower or rack-mount servers that contain one machine per unit. Blade servers consist of a "server on a card," and multiple blades can be mounted in a single enclosure. Each blade has its own memory, processor(s) and storage, but they share a power supply and cooling system. This can lower cost and increase scalability. Blades can be configured in clusters for load balancing and fault tolerance, with blade management software allowing you to add and remove servers without cabling changes.
Blades can also be combined with virtualization software such as VMWare ESX or GSX Server or Microsoft Virtual Server, so that multiple logical servers can be consolidated on fewer physical servers. Virtual machines can be moved from one blade to another easily and quickly, using "hot migration" technology. Either way, adding a blade to the system costs less than deploying a new separate dedicated server.
Scalable network devices
It's also important to keep future growth in mind when buying network devices such as firewalls, routers, switches, etc. In selecting whether to buy a firewall appliance or implement a software firewall such as ISA Server or Check Point, scalability is an important consideration. An appliance may lock you in to a specific hardware configuration. A software-based firewall installed on a PC, on the other hand, can be upgraded with additional processors, more memory, etc. as any other PC can, to handle a heavier traffic load as your business grows.
If you do decide to purchase an appliance, it's wise to budget for a model that can handle not just the number of users and amount of traffic you have now, but what you reasonably expect in the future.
If you're buying a router or NAT device to distribute your Internet service, you might consider a model that has ports for multiple WAN connections, so that if you need more bandwidth in the future you can add a second Internet connection and aggregate the bandwidth of the two (also providing failover in case one of the connections goes down).
Switches should be stackable, so that you can add more as your number of users and/or servers grows.
The key is to always think ahead when making hardware purchases. Don't look just at your current needs, but also at what your needs will be a year or three years from now. The longer your hardware serves you, the lower its total cost of ownership, and you can reduce TCO considerably by making scalability a priority.
Debra Littlejohn Shinder, MCSE, MVP is a technology consultant, trainer, and writer who has authored a number of books on computer operating systems, networking, and security. Deb is a tech editor, developmental editor, and contributor to over 20 additional books on subjects such as the Windows 2000 and Windows 2003 MCSE exams, CompTIA Security+ exam, and TruSecure's ICSA certification.