Designing for High Availability

When it comes to creating high availability solutions, your design plans should not start at the operating system itself ”in this case, Windows Server 2003. To create truly high availability server solutions, you need to start with your hardware first. An important first step in creating high availability solutions is purchasing and using high-quality hardware components from reputable first- tier vendors such as Dell, HP, or Gateway. You also need to give consideration to other issues such as power supply, power conditioning, and environmental conditioning.

Uninterruptible Power Supplies

You might look silly if you go through all the work to create and implement a high availability solution only to have it crippled by a power failure. Uninterruptible power supplies (UPSs) can help prevent this problem. Not only do UPSs provide a backup source of power during power outage , but many of the better UPSs also provide power conditioning . Power conditioning extends the life of sensitive electronic components by removing voltage spikes, voltage dips (brownouts), and line noise.

During a power outage, UPSs automatically switch over to battery mode, providing additional time to safely ”and gracefully ”shut down servers and other infrastructure hardware such as switches, routers, and firewalls. It is important to closely follow the recommendations of the UPS manufacturer when connecting devices to the UPS so that it is not overloaded during a power outage condition, resulting in reduced battery operation time. Many UPS manufacturers also provide software that can be installed on critical servers to monitor the battery condition in the UPS and perform actions such as run scripts or shut down servers as the battery capacity decreases. This graceful securing of servers greatly increases your chances of recovering quickly from the casualty situation.

Fault-Tolerant Storage Systems

Storage systems ”hard drives in common vernacular ”are a key part of any high availability system. Fault-tolerant disk systems can be created in one of two basic ways: either by using third-party hardware solutions or by using dynamic disks within Windows Server 2003. Either way, you should create a Redundant Array of Inexpensive Disks (RAID). There are several different variations on RAID, as discussed in the following paragraphs.

The basic premise of a RAID array is that you can use two or more relatively inexpensive hard drives to create a robust and redundant storage system that can withstand the failure of a disk and still provide disk access. Many hard drives used in RAID arrays are hot-swappable, meaning they can be pulled and inserted into a server while it remains powered on. This feature, as you might expect, can greatly increase server and service uptime by allowing you to replace the failed member of an array without having to shut down the server first.

Although not all-inclusive, the following are two of the more common forms of RAID in use today:

• RAID-1 ” Data is written to one disk and then mirrored in its entirety to the second disk. RAID-1 mirror sets use two identical disks, called mirror volumes . The disks should be the same size. If one disk fails, the system may continue to operate using the remaining disk. Depending on which disk fails, you may experience startup problems and may need to use a boot disk containing an edited version of the boot.ini file that points toward the remaining mirror disk. No parity information is created when using a mirror set.

• RAID-5 ” The entire block of data is written to one data disk. The parity information for data blocks in the same rank is written to the parity disk. RAID-5 requires at least three hard drives of equal size to implement. The total capacity of a RAID-5 array is equal to (N “1)*C, where N is the total number of disks and C is the capacity of one disk.

Figure 5.1 depicts RAID-1 and RAID-5 arrays graphically.

Figure 5.1. RAID-1 and RAID-5 arrays can be used to create fault-tolerant storage systems.

When deciding between using hardware-based RAID and software-based RAID (Windows Server 2003), you should give serious consideration to using hardware-based solutions. When you use a hardware-based RAID solution, you gain decreased headaches as well as increased performance and configurability over software-based RAID. The general problem with software-based (Windows-based) RAID solutions is that you must first reinstall and restore the operating system before you can ever get to restoring data from the RAID set ”something you won't have to face if using hardware-based RAID solutions. As well, when you remove the responsibility for managing the RAID array from Windows Server 2003, the operating system will perform better overall. Also, hardware-based RAID arrays are not limited by the available options and requirements of Windows Server 2003. This allows you greater design flexibility and easier recovery should disaster strike.

Redundant Networking Infrastructure

You should give consideration to making your network infrastructure fault-tolerant. This is usually accomplished by creating redundant routes for traffic to travel over. Also, you can utilize load-balanced or standby infrastructure hardware. Although load balancing and failover are discussed in great detail later in this chapter, a brief description is suitable here. Load-balanced devices divide the total load between two or more devices: If one fails, the remaining "living" devices pick up the load and distribute it evenly among themselves . Standby, or failover, devices kick into action after the primary device experiences a failure condition; thus, the load is transferred over to a new device entirely.

Many infrastructure devices have been designed with redundancy in mind. Vendors such as Cisco have been designing and perfecting redundancy features in their network infrastructure devices for years now. Options such as dual or even triple power supplies and the capability to configure redundancy via the Internetwork Operating System (IOS) itself ensure these devices are no longer a single point of failure (SPOF).