In this system, multiple CPUs and CCUs are sharing and processing tasks, so it is a system configuration with high processing efficiency. There are two types of multiprocessor system. One is LCMP90 (loosely coupled multiprocessors), in which multiple computer systems are controlled by separate operating systems. The other is TCMP91 (tightly coupled multiprocessors), in which multiple computer systems share the main memory and are controlled by the same operating system. The figure below shows an example of TCMP.
92
88 (FAQ) There are many exam questions on characteristics of dual, duplex, and multiprocessor systems. The key term for each system is as follows: “comparing process results” for dual, “switching the units” for duplex, and “sharing the main memory” for multiprocessor systems.
89 (Note) If a failure occurs in the primary system, it takes time to switch to the secondary system. This is because the batch processing or whatever else is being executed in the secondary system must be suspended, and the OS must be booted for the online system. A hot standby system configuration can solve this by standing by, ready to switch at any time. In this case, the OS for the online system stays on, so switching can occur immediately.
90 LCMP (Loosely Coupled Multiprocessor): Each CPU has its own main memory and independent OS. CPUs are joined by a high-speed network or shared path. This is a configuration where independent computer systems are connected via a network.
91 TCMP (Tightly Coupled Multiprocessor): One main memory and one OS are shared in this configuration. Each CPU can perform identical processes, so even if one CPU fails, the processing can continue, albeit with lower performance. This configuration is highly reliable and thus is used in systems where a high level of processing capability is required. 92 MM: Main memory
Switching unit
(Secondary system) Switching unit
2.3.3 Centralized Processing and Distributed
Processing
Points
¾ Centralized processing has a high level of safety but is inflexible. ¾ Distributed processing is economical, but its real substance is hard to
understand.
Depending on how the computers are placed physically, there are two types of processing: centralized processing and distributed processing.
Centralized Processing
Centralized processing is a system configuration in which one computer is connected with many terminals, and the one computer alone does all of the processing. It is easy to maintain the consistency of data, and it is easy to manage the resources. These merits contributed to the popularity of this configuration in which a general-purpose computer is used as the host in centralized processing. Below is a summary of relative comparison with distributed processing.
Advantages of centralized processing Disadvantages of centralized processing It is easy to improve the cost/performance
ratio. (Grosch's Law93)
Operation and maintenance require a smaller staff.
Safety level of the system is high.
Extendability is poor to keep up with new technologies. Backlog can be easily accumulated.94
Overhead of the OS is significant.
Recovery of a host failure is time-consuming. A failure has a far-reaching effect.
Distributed Processing
Distributed processing is a system configuration in which multiple computers connected via a network perform the processing. Since the processing is done through a network, the processing time is longer than that of centralized processing. But, the merit is that a failure of one computer does not affect the entire system. Below is a summary of relative comparison with centralized processing.95
Advantages of distributed processing Disadvantages of distributed processing Management responsibilities are clear. (Management
responsibilities can be delegated to each organization.) Effects of system failure are local.
Maintenance is easy (locally closed).
It is economical as only necessary units are installed
Its real status can be hard to understand. It is difficult to identify trouble spots. Network performance has great impact. Data inconsistency can occur easily. Individual units are managed carelessly.
93 Grosch's Law: It states that “performance is directly proportional to the square of the price.” If the price of a computer doubles, the performance quadruples. However, technological advancement has reduced the prices of devices significantly, so this law is no longer applicable.
94 Backlog: It means systems, software, programs, etc. that are necessary to develop but the development of which has not even begun. The term often refers to those that are held back in the IT department within a company.
95 (FAQ) There are exam questions where you are required to identify the characteristics of centralized processing and distributed processing. For example, questions may be of the form “Which of the following is an appropriate characteristic of a centralized processing system?” Know the advantages and disadvantages of each processing type.
As shown in the following table, distributed processing can be classified according to the distribution status of functions and loads. It is said that vertical load distribution does not exist in reality.
Configuration
Function Function distribution Load distribution
Horizontal distribution Horizontal function distribution Horizontal load distribution
Vertical distribution Vertical function distribution
Horizontal function distribution
A horizontal function distribution system is a system in which computers are classified according to type of application and type of data; examples include processing function distribution and database distribution. For instance, in financial institutions, host computers may be classified into those in an information system and those in an accounting system; this classification is based on the type of processing, so it is an example of processing function distribution. Database distribution means that computers are located in appropriate locations based on the contents of data.
Horizontal load distribution
This is a system in which multiple computers perform processes jointly when an application is executed. When a process is requested, an idle computer responds. In this mode, if one computer fails, the process switches to another computer and is continued. Hence, this system is quite effective in time of failure. A tightly coupled multiprocessor system is an example of this type.
Vertical function distribution
This is a system where the processing function is shared among workstations belonging to individual users as well as computers shared by multiple users. Here, there is a vertical relationship in regard to the processing function. A client/server system is a typical example of a vertical function distribution system.96
96 (Hints & Tips) A client/server system appears as if it were horizontal distribution, but it is properly classified under vertical function distribution. Since one server performs processes of multiple clients, there is a vertical relationship in functions.
2.3.4 Classification by Processing Mode
Points
¾ In batch processing, data is stored and processed all at once. ¾ In real-time processing, data is processed at the moment they come
into existence.
From the standpoint of processing modes, system configurations can be classified into two categories: batch processing and real-time processing. They can also be classified by whether or not they are connected to a network.
Processing mode Operation mode Connection method Center batch processing Offline
Batch processing97
Remote batch processing Interactive mode processing Online transaction processing Real-time processing
Real-time control
Online