Stress testing (sometimes called torture testing) is a form of deliberately intense or thorough testing used to determine the stability of a given system or entity. It involves testing beyond normal operational capacity, often to a breaking point, in order to observe the results. Reasons can include:
- to determine breaking points or safe usage limits
- to confirm mathematical model is accurate enough in predicting breaking points or safe usage limits
- to confirm intended specifications are being met
- to determine modes of failure (how exactly a system fails)
- to test stable operation of a part or system outside standard usage
The term "stress" may have a more specific meaning in certain industries, such as material sciences, and therefore stress testing may sometimes have a technical meaning – one example is in fatigue testing for materials.
Stress testing, in general, should put computer hardware under exaggerated levels of stress in order to ensure stability when used in a normal environment. These can include extremes of workload, type of task, memory use, thermal load (heat), clock speed, or voltages. Memory and CPU are two components that are commonly stress tested in this way.
There is considerable overlap between stress testing software and benchmarking software, since both seek to assess and measure maximum performance. Of the two, stress testing software aims to test stability by trying to force a system to fail; benchmarking aims to measure and assess the maximum performance possible at a given task or function.
When modifying the operating parameters of a CPU, such as temperature, overclocking, underclocking, overvolting, and undervolting, it may be necessary to verify if the new parameters (usually CPU core voltage and frequency) are suitable for heavy CPU loads. This is done by running a CPU-intensive program for extended periods of time, to test whether the computer hangs or crashes. CPU stress testing is also referred to as torture testing. Software that is suitable for torture testing should typically run instructions that utilise the entire chip rather than only a few of its units. Stress testing a CPU over the course of 24 hours at 100% load is, in most cases, sufficient to determine that the CPU will function correctly in normal usage scenarios such as in a desktop computer, where CPU usage typically fluctuates at low levels (50% and under).
Hardware stress testing and stability are subjective and may vary according to how the system will be used. A stress test for a system running 24/7 or that will perform error sensitive tasks such as distributed computing or "folding" projects may differ from one that needs to be able to run a single game with reasonably reliability. For example, a comprehensive guide on overclocking Sandy Bridge found that:
Even though in the past IntelBurnTest was just as good, it seems that something in the SB uArch [Sandy Bridge microarchitecture] is more heavily stressed with Prime95 ... IBT really does pull more power [make greater thermal demands]. But ... Prime95 failed first every time, and it failed when IBT would pass. So same as Sandy Bridge, Prime95 is a better stability tester for Sandy Bridge-E than IBT/LinX.
Stability is subjective; some might call stability enough to run their game, other like folders [folding projects] might need something that is just as stable as it was at stock, and ... would need to run Prime95 for at least 12 hours to a day or two to deem that stable ... There are [bench testers] who really don't care for stability like that and will just say if it can [complete] a benchmark it is stable enough. No one is wrong and no one is right. Stability is subjective. [But] 24/7 stability is not subjective.
Unvalidated stress tests are not advised (such as Prime95 or LinX or other comparable applications). For high grade CPU/IMC and System Bus testing Aida64 is recommended along with general applications usage like PC Mark 7. Aida has an advantage as its stability test has been designed for the Sandy Bridge E architecture and test specific functions like AES, AVX and other instruction sets that prime and like synthetics do not touch. As such not only does it load the CPU 100% but will also test other parts of CPU not used under applications like Prime 95. Other applications to consider are SiSoft 2012 or Passmark BurnIn. Be advised validation has not been completed using Prime 95 version 26 and LinX (10.3.7.012) and OCCT 4.1.0 beta 1 but once we have internally tested to ensure at least limited support and operation.
Software commonly used in stress testing
- IBM Teleprocessing Network Simulator
- IBM Workload Simulator
- Intel processor diagnostic test
- Intel Burn Test
- LinX (AVX)
- Memtest86+ – memory
- Passmark Burn-in
- Prime95, and derivatives such as HyperPi – CPU/heat
In software testing, a system stress test refers to tests that put a greater emphasis on robustness, availability, and error handling under a heavy load, rather than on what would be considered correct behavior under normal circumstances. In particular, the goals of such tests may be to ensure the software does not crash in conditions of insufficient computational resources (such as memory or disk space), unusually high concurrency, or denial of service attacks.
- A web server may be stress tested using scripts, bots, and various denial of service tools to observe the performance of a web site during peak loads. These attacks generally are under an hour long, or until a limit in the amount of data that the web server can tolerate is found.
Stress testing may be contrasted with load testing:
- Load testing examines the entire environment and database, while measuring the response time, whereas stress testing focuses on identified transactions, pushing to a level so as to break transactions or systems.
- During stress testing, if transactions are selectively stressed, the database may not experience much load, but the transactions are heavily stressed. On the other hand, during load testing the database experiences a heavy load, while some transactions may not be stressed.
- System stress testing, also known as stress testing, is loading the concurrent users over and beyond the level that the system can handle, so it breaks at the weakest link within the entire system.
- Destructive testing
- Load and performance test tools
- Black box testing
- Load testing
- Software performance testing
- Scenario analysis
- Software testing
- White box testing
- Technischer Überwachungsverein (TÜV) – product testing and certification
- Concurrency testing using the CHESS model checker
- Jinx automates stress testing by automatically exploring unlikely execution scenarios.
- Highly accelerated life test
- Nelson, Wayne B., (2004), Accelerated Testing - Statistical Models, Test Plans, and Data Analysis, John Wiley & Sons, New York, ISBN 0-471-69736-2
- Sin0822 (2011-12-24). "Sandy Bridge E Overclocking Guide: Walk through, Explanations, and Support for all X79". overclock.net. Retrieved 2 February 2013. (some text condensed)
- Juan Jose Guerrero III - ASUS (2012-03-29). "Intel X79 Motherboard Overclocking Guide". benchmarkreviews.com. Retrieved 2 February 2013.