THE TERMINOLOGY

Whether going it alone or using a hosting service, you will need to have a basic understanding of the technology used. Here are some of more common terms you might encounter during your research.

Processor Architecture

“Processor Architecture” pertains to the overall organizational structure of the computer processor unit (CPU), commonly referred to as the “processor” or “microprocessor.” The main elements of any processor architecture are the selection and behavior of the structural elements and the selected collaborations that form larger subsystems that guide the workings of the entire processor. (For more detailed information on how microprocessors work visit www.howstuffworks.com/microprocessor.htm.)

The microprocessor industry is highly competitive. Consider many factors when selecting your processor architecture including performance, scalability, open/proprietary architectures, on-chip functions, advanced functionality, software availability and, of course, cost.

The Athlon (K7) is a Pentium III-class CPU from AMD with clock speeds ranging from 500MHz to 650MHz. Using a 200MHz system bus, the chip contains the MMX multimedia instructions and an enhanced version of AMD’s 3DNow 3-D instruction set. The Athlon plugs into a slot, known as Slot A, which is similar to the elongated slot used by Pentium IIs and IIIs.

The Hewlett Packard PA-8000 is based upon the older PA 2.0 architecture. This 64-bit processor with a superscalar architecture can execute four instructions per cycle with its two integer ALUs (arithmetic logic units), two shift/merge units, two floating-point units, two divide/square root units, and two load/store units.

The IBM POWER4+ microprocessor is a “server on a chip” that contains two one-gigahertz-plus processors, a high-bandwidth system switch, a large memory cache and I/O. IBM’s POWER family of microprocessors is among the most widely used in the industry and can be found in Nintendo game consoles, Apple computers, and some of the world’s most powerful supercomputers and storage systems.

The Intel Xeon is a Pentium CPU chip designed for server and high-end workstation use. The Xeon plugs into Slot 2 on the motherboard and its L2 cache runs at the same speed as the CPU. Xeon introduced the System Management Bus (SMBus) interface, which includes a Processor Information ROM (PIROM) that contains data about the processor and an empty EEPROM that can be used by manufacturers to track their own information such as usage and service information. Xeon chips can address 64GB of memory.

The MIPS Technologies R10000 Microprocessor is a four-way superscalar architecture capable of executing four instructions per cycle, which are then appended to one of three instruction queues (integer, floating-point, or address). Each queue can perform dynamic scheduling of instructions. Although instructions are executed in order, they don’t need to be, which allows the R10000 to maintain up to 32 active instructions at a time that are in the process of being executed. It has five independently operating execution units (2 integer ALUs, 2 floating-point units, and a load store unit for generating addresses).

The Sun Technologies’ UltraSPARC (Scalable Processor ARChitecture) is an open reduced instruction set computer (RISC) specification to which anyone can build compatible chips and in which a microprocessor is designed for very efficient handling of a small set of instructions. The UltraSPARC architecture scales very well, ranging from low power notebooks and portables to multi-million dollar Cray scientific research supercomputers. The UltraSPARC based systems dominate the UNIX server market and provide the processing power behind many of today’s robust web-hosting services and Internet Service Providers (ISPs).

Clusters

One way to surpass the limitations of a single server is clustering. Clusters are essentially multiple servers that are used as a single unified resource through the use of software, switches and routers. The inter-connection of multiple servers sharing resources results in greater availability due to the ability of the other systems in the cluster to assume the workload of a failing resource.

Maintaining access to data is a key element of the concept of high availability and if your server is running but can’t reach critical data, your web-based business can come to a standstill. Clusters depend on: strong data sharing models, systems that can co-exist at different software release levels, the ability to dynamically vary systems off-line, and robust recovery models that are able to keep everything up and running. A cluster also must be able to perform load balancing. This allows processing to be evenly distributed among the various machines that comprise the cluster. In extreme cases, the software will automatically exclude a failed server from the cluster, allowing the working servers to take up the slack.

To the network’s devices and your administrator, the cluster appears as a single server. This streamlines your network management efforts.

Load Balancing Switches and Routers

Other load bearing solutions include load balancing switches and routers that distribute traffic to a group of servers, sharing the load among them.

The first versions of load balancing routers were designed specifically to support Internet traffic, especially web servers. They used a round-robin algorithm, distributing requests to each server in sequence. Newer load balancing router algorithms provide a more even load distribution across a group of web servers. They react to the traffic coming through the load balancer and distribute the traffic according to the load on each server. Some balancers can check to see if a system request is for data that is residing in a server’s cache, thus making it easier for that server to respond quickly. They also provide monitoring of the actual load level on each server in real time, allowing each server’s load to be kept perfectly balanced.

In a server farm configuration, the servers are usually identical computers that are large enough to handle only a fraction of a website’s total traffic. The number of servers needed varies. Some web operators opt for a few large servers and others choose several smaller, less expensive servers. With either option, the load-balancing algorithm can be a simple round robin, or may be more sophisticated, taking into account each server’s current load. A server farm configuration provides high availability since multiple machines can take on additional processing in the event of the failure of any single server. This type of server configuration is cheaper and easier to implement than a high availability cluster since the servers don’t need to be aware of each other. Only the load-balancing component is aware of all of the servers. In the event of a failure, there is no complex failover process. Instead, the system just stops sending requests to the failed server and routes traffic to the remaining servers. In this way, you get both availability and scalability.

Web server load balancing produces a cost-effective way for websites to respond to growth spurts without the need to replace equipment every few months with a new, higher-capacity system.

Rack Units

A rack unit is a vertical shelving system to mount servers. Racks can be freestanding or else bolted into the floor or wall. Racks generally hold rackmountable computers that are 19 inches wide and have faceplates that allow the computer to be affixed to the rack’s frame via screws. Two or more rackmount servers can comprise a high-availability groupware (sharing data across a distributed system) and web-server system. Mounting servers in racks helps with space problems and allows for ease of management when it comes to the formulation and assignment of security measures and redundant power supplies.

Generally each “layer” of a rack holds one 19-inch rackmount computer, but smaller computers from companies such as Crystal Group, Swemco and APPRO allow for up to four or more PCs to share a shelf on a rack.

Multiple-CPU Servers

A multiple-CPU unit is a good choice for websites that demand high-end computing environments. It is a viable option used to satisfy the server consolidation requirements of very large websites that offer, in addition to products and e-commerce options, services such as free email, chat rooms, streaming video, etc.

The choice of a multiple-CPU unit should take into consideration the need for the flexibility in mixing and matching components in a rack. Choose multiple-CPU units if you are unsure of exactly what your needs are now or what they might be in the future.

Another advantage of multiple-CPU servers is that redundant power supplies are among the basic requirements (not an added feature) for an enterprise class system such as the 8-CPU Intel Servers. Consider, however, that these multiple-CPU units can cost in excess of $20,000.

Server Cabinet

A server cabinet is a metal cabinet designed to house rack-mounted servers (some also house tower configured systems). A good server cabinet will usually have a slotted front door, a perforated steel rear door and a perforated top panel to assure maximum airflow. There is room for fans or blowers to be added, which are usually necessary. The side panels usually lift off for easy and quick component accessibility.