Computer performance

The computing power (also called English computing power or performance) is a measure for calculating machines and data processing (short- DV systems) or information technology systems ( IT for short ) systems. As a rule, is the data processing speed ( for the calculations per unit time), colloquially known as short speed or quickness, machine parts used (such as the or the central and graphics processor units), and the speed of individual applications (such as z. example, simulation calculations or processing large database administrations ) in the center of attention, and rarely also the computing power of entire IT systems, such as mainframes, or composite thereof. supercomputers in so-called

The data processing power precision (also called performance) - - and the physical performance of an IT system describes the work done or the amount of energy used per unit of time is considered Moreover, with the performance.

• 4.1 Measurement
• 4.2 forecast

Additional Selections

In terms of " how fast" performance, as long as the computer system for executing a processing order needs. Such an order may be about the online response to a mouse click or the settlement of a large background job. " How fast " but can also be the question of how many jobs can do the DP system per unit time. This is the job throughput.

In addition to this description, the term is also used for other properties such as functionality, energy efficiency or reliability.

Components and system performance

Component performance

Component power means the power of an individual component of a data processing system, such as the processor, the main memory or the network connection. The user community is a component of the surrounding devices shaft, such as the machine- command generating environment of the processor, the memory- access -forming environment of a storage system or the data transfer orders generating set of nodes in a computer network. The DV - performance components to describe with performance measures such as distribution of job completion time, throughput rates of orders or mean response time. Where appropriate, such variables are still differentiated by the type of contract, for example, read and write requests in memory systems. With components power the set of all output sizes is referred to in this case. To assess whether the component has a satisfactory performance, a review is carried out, the opposite is the determined power values ​​to the user community (ie the mechanical environment of the component ) values ​​required by the selected performance measures.

System performance

System performance with the performance of a complete data processing system will be referred to, which may consist of a plurality of components. Corresponding parts of computer systems, individual software components ( application software and operating system), computer ( for example, file server ), computer networks (for example, computer clusters) or special devices (for example switches). Furthermore, it depends on how efficient algorithms are implemented and how the software and hardware is configured.

It has been repeatedly tried to calculate system performance from the known components services. Experience shows that the correlations are generally too complex, so this does not happen. Reliable system performance values ​​can only be obtained from measurements or with restrictions, from predictions.

Characteristics and performance criteria

There are a large number of parameters, which are used for performance evaluation. In many cases, these figures tell only little about the actual performance of the system.

• Instructions per cycle (IPC )

• Instructions per second

• Floating Point Operations Per Second

• Data transfer rate

• Data throughput

• Response time

• Response ratio

• Frames per second

• Processor clock

• Latency

• Access time

Measurement versus prediction

DV output is described with output sizes. For the determination of numerical values ​​of such quantities there are the following ways:

Measurement

Measurement (English measurement) is the experimental determination of DV performance values ​​with the actually constructed computer system. The supplied to the system job stream may be generated by real users (real load) or by a user simulator ( simulative load). The simulative load may be an individual load mostly in the context of a load test or a standard load for a benchmark comparison.

A distinction is made between software and hardware measurement measurement.

• Hardware measurement

• Software measurement

Forecast

Prediction is a procedure in which the numerical values ​​of DV output sizes are determined by mathematical and analytical methods or by means of simulation methods. Unlike the measurement methods of calculating performance evaluation must be present no real system.

In the analytical method, the DV system and its users are represented by a mathematical model and the DV output values ​​determined mathematically. In the simulation process, both the DP system as well as its users are simulated and determined from this simulated events, the numerical values ​​of the DV output sizes. Both the analytical methods as well as in the simulation process, the results are only approximate or estimated values ​​in general. They have the character of a forecast or prediction in contrast to values ​​from measurements with a real computer system.

• Graph -theoretical description

Especially in communications technology, it is advisable to model the system as a graph. The components are represented as nodes. Connections between the components are represented as edges. Each edge has a maximum capacity which must not be exceeded and a current flow. The resulting network can be evaluated by the maximum flow between two components is determined now. If the pairs performed for all nodes, so can locate slow components.

• Traffic Theoretical Model

A transport theoretical description is based on a queue with orders which will be processed by an operator station. The orders reach the queue with a mean arrival rate and leave the control station with a mean service rate. The traffic intensity is described by the quotient. The system works only reasonable, is long. Otherwise it comes to overflow.

With this model different systems can be represented. A computer without pipelining with only one processor gets a negative exponential distribution as a distribution function. Pipeline processors with steps are modeled using the Erlang distribution. For multiprocessor systems, one uses a Hyperexponentialverteilung th order

Queuing model for pipeline proze

Queuing model for multi-processor systems

A further attempt to compare the performance of processors analytically are mixes and core applications. These attempts, however, are relatively expensive and are rarely used today.

• Befehlsmixe

• Core programs

Review of DV - performance

Performance values ​​(whether determined by measurement or by prediction methods ) are numerical values ​​of physical quantities that are important and interesting, still do but seen not in itself testimony to the important question whether the DV system, the (power) needs of its satisfied user community. These needs must therefore be numerically defined.

Second, it should be carried out a comparison of these ( required ) values ​​with the services provided by the DV system performance values. The result of this comparison is the statement that the DV system is sufficient to meet the user needs. This is the review. Thus we come to the final users of the benötigtem result of "insufficient", "sufficient", " exceeded " etc. The scale of such Endaussage could also be selected in more detail.

It should be noted the following items:

• The values ​​of DV output variables of the system under consideration are concrete numbers. You would only change if the system would be changed (for example, through the exchange of hardware components such as the processor or memory unit or software components, such as the operating system version and / or application software with other software efficiency).
• In contrast, the evaluation results of the user population, the rating by reference, depending. For example, the assessment of a considered computerized system for the user community A turn out very well, while the same system is unsatisfactory for the user community B.

Benchmark and load test

While the goal of load testing is to provide evidence to whether the anticipated load can be processed in the required time, it is the goal of the benchmark to determine a measure that can be compared between different systems.

The load test may be designed so that the load current generate real user (real load). The load current can be but also by a simulator, which simulates the entire user population detail, produces ( simulative load). A benchmark always used a standardized, simulative load the results to compare.

Must To achieve more accurate results, a measurement software are used exactly which logs the job action and after completion of the test performs the evaluation ( determination of the DV output sizes ).

Over time, a tremendous amount of (computer) benchmarks has been developed and described, both at the scientific level, as well as industry and the economy. These benchmarks have almost all the different principles and DV output sizes, so that measurement results are generally not comparable. Many of these benchmarks had only a short timeliness and are gone.

Performance databases

Journals and magazines regularly publish rankings on the performance of computer systems or components. These are determined by indicators or benchmarks.

A well-known database, the TOP500, lists the 500 most powerful supercomputers in the world. To this end, the Linpack benchmark is used.

Standards

Add to full proposals to the benchmark methodology and DV output sizes make the standards DIN 66273 "Measurement and evaluation of the performance of computer systems " and ISO 14756 "Measurement and evaluation of the performance and efficiency of the software of computer systems ." The ISO standard has adopted the principles of DIN 66273 and expanded. Measurements that have been carried out in accordance with DIN 66273, are also compliant with ISO 14756th The ISO standard expands the application field of the measurement and evaluation of DV - performance addition to the measurement of (term ) efficiency of system and / or application software. The DIN standard standardizes the benchmark methodology, omitted because of the ephemeral nature of benchmarks but on defining concrete benchmarks. However, the ISO standard also contains more examples of complete benchmarks.

The Application Response Measurement (ARM ) is a standard of the Open Group and is used for measuring the performance of transactions from the user's perspective.

Variable power

In general, the performance of an information technology system is constant. However, for backward compatibility, or, to save energy, a reduction of the power may be appropriate. The turbo button made ​​in 8086 to the PC Pentium era for backward compatibility with the IBM standard. This is often done by reducing the clock, as well as by shutting down the level 1 cache, or a reduction in the clock frequency of the front side bus.

In contrast, modern notebooks technologies such as PowerNow!, Cool'n'Quiet or Intel SpeedStep technology to reduce power when not needed to conserve scarce energy resources in the accumulators. This can be done by reducing the clock or the core voltage or shutdown of individual processors in multiprocessor systems.