New Page 1

Clustering, Scalability, and High Availability

These concepts are not so easily understood as may appear on first glance. You’ll notice clustering, scalability, and high availability are not separate, distinct concepts, but rather are closely interrelated.

Clustering

A combination of hardware and software solutions used to produce a group of redundant, mirrored servers in a network that will take over quickly if the primary device fails or is shut down. This is protection against various types of disasters.

Use for Load Balancing and high availability, especially to achieve 99.999% availability (called the five nines; see below).

An optimally designed cluster should appear, to the user, to be a single system.
Scalability

If your cluster is designed properly, it will be able to change and grow. This ability to continue to function beyond its original configuration is called scalability.

Scalability upward, rather than downward/backward is the norm. Scaling downward usually means trying to achieve the same goals in a more constrained environment, which is not a good practice.

(Note: On the Network+ exam, if something, anything, is not a good practice, then it’s a wrong answer.)
High Availability

This the percentage of time over the period of a year that a network is available, or falls within an acceptable range (otherwise the availability would not be high).

Optimally, the percentage should be between 99.9% and 100 percent (note: 100 percent is not possible, but should still be a goal). The most often mentioned percentage, in IT, is 99.999 percent, AKA, the “five nines.”

Five nines works out to 5.39 minutes of total downtime—planned or unplanned—in a given year. Issues that affect the amount of downtime include:
- Continuous Availability: The customer expects 24x365 non-stop service with no interruptions. This is the ideal state—high availability is not defined as continuous availability.
- Fault Tolerance: Striving for continuous availability, even when hardware and/or software fails. This is accomplished by having redundancy in hardware, including CPU, memory, and I/0 subsystems. This is another ideal state because high availability does not imply complete fault tolerance.
- Single Point of Failure (SPOF): If the loss of a single hardware or software component results in the loss of service, it is a SPOF. These are components that are not backed up by redundant replacements, and should be. Fail over: Eliminating SPOF with redundant components that will take over automatically should the primary component fail.

Three ways to achieve the five nines:

Establish systems designed for high availability. Many components within the system must be duplicated for fail over. However, this can be expensive as it would require redundant components that seldom if ever are used.
Establish mirrored systems, so that if one fails, the other takes over.
Implement clustering, in which matched components do not necessarily have to be duplicates of each other, but act as partners to share the load. One of the suggested ways of linking a cluster is through Fibre Channel (see below).

For excellent articles on the five 9s, click here and here.
For the IEEE Task Force on Cluster Computing, click here.
For a free Web-based training class on High availability, click here.
For basic troubleshooting advice, click here.
For an article that explains why you shouldn’t be afraid of the command prompt, click here.
For the hows and whys on the distribution of resources on a network, click here.

UPS (Uninterruptible Power Supply)

Nothing on a network will work without electricity. The sudden removal of electricity not only keeps the network from functioning, but also can cause a great deal of damage to hardware and software. There are also spikes, surges, and other uneven electrical power flows, which also can cause damage. To counteract these electrical forces, it’s always a good idea to install UPSs.

A UPS is a device that keeps the electrical flow to the network device steady, free of spikes and surges, for a short time when primary electrical power is lost. A UPS avoids any break in power flow by routinely providing power from its own inverter, even when there's no interruption in service. The switch to battery power as soon as the UPS detects a loss of primary power gives you time to save any work you’re doing and gracefully shut down the device, such as a network server. A UPS also provides protection from power surges, spikes, and other power anomalies.

Don't confuse a UPS with a Standby Power System (SPS). An SPS monitors the primary power supply and will switch to the battery power once it detects a problem. The difference is that an SPS takes time to switch over. The SPS can take several milliseconds, during which time the network device will receive no power. Another term for an SPS is “Line-Interactive UPS”. UPSs are generally much more expensive than SPSs.

How UPSs work
APC’s UPS page