It's hard to get the best use out of a tool (such as an automobile) without understanding how it works. With that observation in mind, we attempt here to define the various pieces of a workstation as well as give some of their general speed and size requirements. The workstation is a computer small enough in size and cost to be used by a small group or an individual in a work location yet powerful enough for large-scale scientific and engineering applications. Typically the workstation employs a Unix operating system and has good graphics capability.
Although the specific speeds, sizes, and components of various workstations are constantly changing, the basic ideas should remain the same. Listing the minimum configuration needed to run Unix well, however, is not a clear-cut matter, since it depends on what the system will be used for and what performance will be expected from it (analogous to the engine requirements of an automobile). Users with ``hot'' systems generally cannot imagine enduring the pain of living with anything less, yet acquiring even a minimal configuration is a great advance for those with inadequate systems. In an additional analogy to the automobile, how fast a CPU or how large a block of memory you purchase will often be determined more by your budget and ego than by the requirements of the software. Even if you try to be Spartan, the new components introduced every year as optional luxury items often become required in the future. So you may well end up buying them but at a time when they have become less expensive.
The collection of programs known as Unix has grown a lot in power and
extent in recent years. This growth, combined with the wide use of
CPU-intensive graphics displays and their associated use of the X
Window System, means that a Unix workstation requires a fast
processor.
It is easy to benchmark a particular numerical program to
see exactly how long it takes a certain machine to turn out the right
answer. It is harder, however, to determine ahead of time what the
response time of your computer will be in an actual working
situation when there are a number of users, a number of programs
running in the background, and so forth. Since user productivity is
significantly affected by this response time,this is an important
consideration.
Let's say your usage includes a good number of X applications, X terminals, graphics packages, symbolic manipulation users, and CAD/CAM users. You may find the response time of the system, and thus its usability and everyone's productivity, increase greatly as you move to a faster machine or a network of machines.
As CPU speeds and the scope of computations have
increased, this limit has become a hardship for some users and may
well be changed in the future.
Hard disks are rated by size and average access speed. The size is usually stated in megabytes for an unformatted disk. You can expect a 15- to 20% decrease after formatting. The smallest (PC) Unix systems require a minimum of 100-150 MB, while a workstation Unix may well require 300 MB minimum. While a 600 MB hard disk is adequate for scientific work, it is not uncommon to see systems with more than 1GB of disk space.
The access speed for a hard disk is the average time it takes the disk to move its ``heads'' to a place where it can read or write your data. Currently slow drives, such as found on laptops, have access times of 50 milliseconds or more, while fast workstation drives have access times of 15 milliseconds or less.
.
The console is the
communications center of the system while booting, the bulletin
board for error and informational memos during operation, and should
be the place where the system administrator works during system
maintenance. From a systems point of view, workstations don't require
a graphics display but do require a ``console'' as a device to handle
text information. Depending upon the workstation, this could be a
simple serial terminal or a small monochrome display.
Another attraction of SCSI is that it allows users to add hard disks or tape drives which are external to their workstations. This means the user can just plug a cable into a port on the backside without having to handle the ``guts'' of the computer. The external device resides in its own case with its own power supply and can be repaired or replaced with minimal disturbance to the system (and since drives are mechanical, they will break down regularly). While this may suggest to you that you need a computer with many ports on the back, that is not necessarily true since additional SCSI devices can be connected to the extra port usually located on a SCSI device. Having a number of such devices connected to each other with cable loops is called a daisy chain. At present, seven devices are the maximum number which can be controlled by a single SCSI bus.
Although ethernets are fast compared to PC networks like Appletalk, they can transmit only one message at a time, a so-called packet. The packet contains the data to be transmitted as well as the address of the destination and a return address. As the number of machines on a network increases, the fact that they all must share this 1 MB bandwidth means that the effective bandwidth for any one computer is decreased since it may have to wait while another machine's packet gets delivered. These packet collisions do not become much of a performance problem until there are many machines on the same network or until the network is used heavily for activities such as Network File System (NFS). If the traffic does become a problem, it may be a good idea to break the network up into subnets joined by routers or bridges .
The physical characteristic of your workstation's ethernet is either ``thin ethernet'' or ``thick ethernet.'' Both carry the same electrical signals, yet do so on different size coaxial cables and with different connectors. The thin ethernet cables are lightweight, 75-ohm coaxial cables that get looped from one machine to the next. With thick ethernet, a 15-ohm cable connects the computer to a transceiver box which is securely fastened to a thick coaxial cable. The advantage of thin ethernet is lower cost and ease of cable hookup. The disadvantage is that if any one connection to a machine is broken, the entire network is down. Further, thin ethernet supports fewer connections and can be extended a shorter distance (185 m and 30 nodes for thin ethernet and 500 m and 100 nodes for thick ethernet). Commonly, a local group's ethernet may be thin while a department's ethernet may be thick.