Performance Test Principle - Throughput Analysis

  Many people are not very clear about the concept of throughput. Some people think that TPS is throughput, and some people regard the amount of data processed by the system per unit time (kilobytes per second) as throughput. So which is the correct answer? TPS (transaction per second) ), as the name suggests, is the number of things processed per unit time. The standard for throughput with TPS is rather vague. Because of the size of a transaction, the complexity of the transaction varies. This formulation was mentioned on Jmeter. It focuses on application scenarios and is closely related to business services. It is more appropriate to measure the throughput of network equipment with the amount of data transmission per unit time as the standard. In performance test analysis, two methods are often used together, one is to meet the measurement needs of the application, and the other is for the analysis and design needs.


      The term throughput existed before the birth of website services. So what was the throughput at that time, and what kind of scenarios was it applied to? Port throughput refers to entering, The quantity of goods that have been loaded and unloaded out of the port area. Also known as port throughput. It is the most important indicator to measure the size of the port. Reflected in certain technical equipment and labor organization conditions, the number of cargoes loaded and unloaded by the port for ships within a certain period of time, expressed in tons. When the network enters people's lives, the throughput of the network is also used to represent the transmission capacity of the network. Because the network passes data (data), then the throughput of the network is naturally measured by the size of the data transmitted per unit time. The same is true for the throughput of the server, which is expressed by the data processing capacity of the server per unit time. In this sense, the throughput of the port is the same as the throughput of the network equipment and the server.


      If a port design throughput is 260,000 tons. 5 ships of 10,000 tons are handled respectively. There are 10 thousand-ton ships and 200 hundred-ton ships. Handling loading and unloading capacity per day, 10,000-ton rounds, 2 thousand-ton rounds, 3 100-ton rounds. That is: ((5x10000x1)+(10x1000x2)+(200x100x3))x2=260000 tons. If the ships entering and leaving the port must be distributed in and out of the port according to the design, then the port can handle 260,000 tons. But if all the 10,000-ton ships enter the port, the berths of the 10,000-ton ships are waiting in line, so only 5 ships are loaded and unloaded every day. Obviously we can't say that the port is only capable of handling the throughput of 5 ships? But if the schedule is designed to bring different sized shipping vessels into the port each day, then the design throughput can be achieved.


      Similarly, if the throughput of a network server is measured by TPS, it is undoubtedly ambiguous. The size and implementation of a transaction are very different, which determine the processing time and resources, and the TPS varies. It is not difficult to draw out that it is unscientific to use TPS as the standard for throughput. However, from the perspective of business services, users often only have a single and few demand scenarios, and sometimes use transaction processing capabilities as throughput standards, service agreements, and SLA standards. For example, a transport fleet only has 10,000-ton ships and no small boats. The service agreement between the transport team and the port will be based on the guaranteed maximum throughput of 5 ships per day. But it does not mean that the port only supports a throughput of 100,000 tons. In the SLA of the website application service, the transaction processing capability can also be used as the standard. For example, the number of logins the server can support per second.


      In the process of development and performance testing, we usually focus on two indicators related to throughput, one is TPS and KiloBytes/Sec. TPS is the performance capacity indicator of application services. The second is the measurement of performance capacity in terms of data processing capability. The second measurement method is especially suitable for the measurement of server capacity before designing application products. The output throughput can be used as a reference in the design of application products.