Edit this article: Reiko
Server stability is the most important, if we can not ensure the needs of business operations in terms of stability, then a high performance is useless. Formal server product vendors will be running stability tests at different temperatures and humidity. Key consideration is redundant functions, such as: data redundancy, network card redundancy, power redundancy, redundant fans and so on.
Some server stability test method is mainly for the following:
pressure test:
The number of known system peak period use, verify the number of each transaction concurrency (through the peak number of conversion) under the transaction response time can meet customer requirements. Each performance index system in which pressure is still within normal values of. Whether the system will lead to such pressure because of adverse reactions (eg: downtime, application abort, etc.).
Ramp Up incremental design:
Such as 75 concurrent users, the system as a registered user 1500, 5% to 7% as a reference value of concurrent users. Generally manner every 5 people 15s pressurizing load design, the main reference value press machine performance testing, several suggestions Run. With transaction rate and error rate measured by the actual loading.
Ramp Up incremental design goals:
Looking incrementally been pressurized system performance bottleneck location, inflection point performance to seize the opportunity arise, commonly used reference Hits hits and throughput, CPU, memory usage comprehensive judgment. The number of analog peak period use, such as the morning login, quit after work, wages message system when sending.
Another extreme simulation mode, can be regarded as the peak pressure in the case of the operating system while clicking on the transaction limit of operation indicator. Constant pressure mode, set the same name in the script rendezvous point in the transaction (such as: lr_rendzvous ( "same");) in scene design, the use of a transaction point set policy. At the same time in order to achieve a collection percentage point as a standard, while the release of all is Run the Vuser.
Stability Test:
The peak number of known systems, the operating frequency of each transaction. Design of integrated test scenario, test each scene according to a certain proportion of run together, simulated users are using the situation for several years. And monitoring tests, performance of each system in which pressure is to maintain normal values. Fluctuations or whether the transaction response time increases with time over the test will appear. Whether the system such as downtime, application and other abnormal abort happens over a period of testing.
According to the above tests, the performance of the position of the inflection point occurs in each transaction conditions to determine the stability of the test the number of concurrent users. Based on the actual test server (pressure machine, application server, data server tripartite performance), estimate the final number of concurrent users.
Can also be tested in the following way server, verify that the server with automatic handling mechanism in special circumstances:
1 fault tolerance test
By simulating a number of abnormal conditions (such as: Server suddenly power outages, network intermittent, insufficient server hard disk space, etc.), verify the system in these situations whether the automatic handling mechanism to guarantee the normal operation of the system or restore operation measures . If HA (automatic disaster recovery system), but also specifically for these additional test automatic protection system to verify its ability to effectively trigger protective measures.
2 Troubleshooting Test
, Verify a test module for the existence of hidden problems have occurred or are suspected by the original system or experience to judge the case, to verify whether these modules will happen the same performance problems. Such as: Upload attachment module memory leaks and address book module optimization, open the Tivoli performance monitoring on the impact of OA system performance and so on.
Assessment tests are used to test the relevant key performance indicators point acquisition system carried out. Mainly for pre-test results no clear expectations, but to obtain performance under certain stress scenarios (such as: transaction response time, number of concurrent users) pass the test.
Evaluation transaction response time: test activities to obtain a transaction carried out in a specific response time under pressure. By simulating various known clients peak pressure value or expected value can withstand pressure, obtain transaction response time under this pressure.
Reviews affairs of concurrent users: test activities to obtain a transaction can withstand the number of concurrent users in a particular system environment carried out. By simulating the real environment or directly from the real environment, the evaluation in this environment can withstand firm concurrent users. Standard threshold value is determined to be pre-defined (e.g., response times, CPU usage, memory usage, CTR has peaks, peak throughput has occurred, etc.).
Evaluation system of concurrent users: the number of concurrent users for access to the entire system can withstand the test carried out activities. By using the ratio of the major modules and frequency of each analysis item beforehand, defined as the ratio of the percentage of each transaction in the integrated scenario, each transaction in a manner proportionate allocation of the number of concurrent users. Simulate real-world or directly from the real environment, the number of concurrent user reviews in this environment the system can withstand. Predefined threshold criteria (e.g., between response, CPU usage, memory usage, CTR has peaks, peak throughput has occurred, etc.). Value Standard to barrels law prevail (the smallest number of concurrent concurrent transaction for the entire system).
Affect the amount of data in different databases on performance evaluation: a database for different amounts of test data, the test results were compared, analysis showed that influence the amount of data in the database tables of transaction performance. It is determined in advance after long running, or some module customer risks that may exist a large amount of data.
By the above test or actual operation of the user it has been found that the system performance problems or suspected performance problem exists, in response to the need to reproduce the test scenario or question of definition. If possible, you can directly identify the code or module cause performance problems are located. Such tests, mainly the problem by testing the script scene, and can increase the discovery and testing tools such as open Tivoli performance monitoring, open HeapDump output, Linux resource monitoring commands, supplemented by manual testing at the scene during the operation.
Professional services, professional attitude from.