Distributed Processing — Past, Present and Future

by Lloyd Borrett
Technical Cornucopia, March 1991

Remember distributed processing back in the late 1970s? Intelligence at the workstation was primitive back then. All of the benefits of distributed processing were based on multiple host computers, each with their own set of input/output devices.

One benefit of distributed processing in the '70s was lower processing costs, based on the fact that a number of networked minicomputers could be acquired and maintained at a lower cost than running a centralised mainframe. Another benefit was reduced downtime. If one host went down, then processing could be distributed among other host computers in the network.

At the time, more efficient operations were promised, since it was asserted that processing could be performed at the location of the business transactions. Even a more stable staff of MIS professionals was promised, since people in remote areas could become an integral part of the information systems effort in organisations.

Now, if these notions sound primitive, or to be more kind, a little rough around the edges, one must remember that distributed processing over a decade ago didn't gain the notoriety that was heralded in the media at the time.

Intelligent Workstations

What was missing? Intelligent workstations — machines that facilitate the use of fourth generation languages, graphical interfaces, and CASE tools, allowing the power of computing to reside within the control of users.

The first attempt at distributed processing can be summed up as follows: You can take an autocracy and replicate it. What you end up with is more than one autocracy.

The concept of user "freedom to compute" has made distributed processing legitimate, and intelligent workstations are required for this concept to become reality.

Just what is an intelligent workstation? In the scientific and engineering community, intelligent workstations are often high-powered 32 bit standalone units, capable of processing mathematical computations at great speed, while at the same time displaying high-resolution graphics. They usually run a multitasking operating system, such as UNIX or one of its many derivatives, and generally allow for interactive screen control. Workstations also support very large amounts of RAM and disk storage. Increasingly, high powered DOS based personal computers are also being used as traditional workstations.

This is the classic definition of a workstation. These characteristics, as well as supported computing activities, are independent of traditional host computing activities.

Host Computing

The classic scenario for computing in a host environment is the use of dumb terminals that may communicate with one or more host computers. Generally, there is no intelligence in the terminal that can be accessed and manipulated by the user. These terminals usually execute one program at a time. They run under a multitasking operating system in a shared processor using shared memory. A negative aspect of this environment is that if one program loads the processor or memory, the performance of all programs running on the system can suffer.

An accepted definition of an intelligent workstation is this: A device that provides processing capability in the local input/output device. An intelligent workstation addresses one or more CPUs, thereby taking the characteristics of a communications platform. One of the CPUs is dedicated to the workstation itself, whilst the other CPUs are on remote hosts or servers.

Intelligent workstations can be described as single or multiuser multitasking devices that support active and passive intelligence attributes.

An intelligent workstation implies a multitasking environment. Multitasking activities may be supported by the local CPU and/or remote processors.

In a limited sense, a host terminal emulation card (i.e. 3270 emulation card) installed in a PC allows simultaneous access to multiple host sessions. Sessions not currently in view by the user may be running procedures on the host while the user is tending to other host sessions or running applications on the PC. Here, the multitasking environment is supported by a remote CPU, while the local CPU (microprocessor) is managing the emulation software as well as other programs and data in the local device. This local processing may itself be managed in a multitasking environment.

An example would be to run both a local database and a spreadsheet at the same time under Windows, on a PC based workstation. Another partition would support the resident terminal emulation software.

An intelligent workstation is generally a single-user device, allowing one user to operate the input/output (keyboard/display) functionally at any given time. However, the local CPU, memory, and storage devices may be shared by other workstations, as in the case of a non-dedicated LAN file server.

Active and Passive

Intelligence can be described as both active and passive intelligence. Active intelligence includes CPU cycles that are under direct control of the workstation user. Personal computing activities are the prime example of active intelligence.

Passive intelligence allows CPU cycles in the workstation to be used by other CPUs in the network. A workstation can be a communications server, a print server, a file server, or a database server. All can take part in network activities independent of active intelligence applications, and most times without the knowledge of the local user. A variation is the dedicated database server. When coupled with a set of active intelligent devices, one ends up with a platform that supports the client-server model.

The functionality of intelligent workstations should be designed to be consistent with the requirements of the applications they'll support. External influences, such as vendor strategies, will also impact the functionality of intelligent workstations, providing a significant influence on their design.

Workstation products must conform to connectivity methods, which aren't always consistent between vendors. One of the reasons that the PC has become so popular is the fact that it has an open bus, and can be made to comply with different vendors' connectivity policies through the use of add-in board level products. Various LAN cards can be installed, allowing the PC to attach to Ethernet, Token Ring, or other types of LANs. Host emulation cards can be installed allowing the PC to connect to various host computers. The lowly RS-232 port, available in almost all PCs as a standard connectivity device, can allow a PC to connect to a large number of different computers.

Workstations can be diskless. A diskless workstation is a desktop computer with no local disk drives. It reads required data from the hard disk of a shared server. Since a diskless workstation has its own CPU, it is by definition an intelligent workstation. If the CPU is located in a unit that includes support for the monitor, keyboard, RAM, video interface, and I/O interfaces, then it is designed to be connected to a LAN. If the CPU and other supporting components are located remotely from the monitor and keyboard, then the diskless workstation is supported by a clustered CPU architecture.

An intelligent workstation may or may not be a diskless workstation, and a diskless workstation may or may not be an intelligent workstation.

Distributed processing is directly related to intelligent workstations in the modern day era of network computing. This relationship has fostered models such as the client-server database model. This model, in turn, has fostered the introduction of application development tools designed to optimise the distributed processing power of today's networks.

Intelligent workstations and distributed processing are the cornerstones of network computing. In fact, network computing is about to replace host-based computing as the de facto standard of computing models. If the cornerstones are in place, we must examine the bricks and mortar required to build upon them. These are the server platforms, operating systems and application development tools.

Choices

However, choices must be made. A server platform must be considered. Should it be the more traditional minicomputer or the new genre of fault tolerant file servers? Which network operating system software should be used: OS/2 LAN Server/Manager, LAN Manager/X, Novell Netware, Appletalk or NFS.

Finally, which application software tools should be utilised? Some organisations have been doing nothing but waiting for the promised client-server software. Others have been distributing data processing across networks for more than five years.

Today, as we enter the '90s, the important issue is how we can realise the benefits of current and future technology. The solutions lie in the concepts of intelligent workstations, distributed processing, and network computing.