An “Action” in the XLT programming model represents a single, enclosed step in a test case. Actions are blocks of code that can be reused in several test cases. An important action for testing a web search engine could be to enter a search phrase and then click the search button to list the results. An action can trigger one or more requests.
An agent simulates a number of virtual users repeatedly running certain test cases against the system under test.
Agent Controller / Agentcontroller (XLT)
On every load generating system there is at least one
running which receives command from the master controller and sends back
test results. The agentcontrollers (we often use the abbreviation “AC”/“ACs”)
control local agent processes generating the load.
To be better highlight that an
agentcontroller is a shipped XLT component, we prefer the spelling agentcontroller over agent controller but you might see both spellings used.
An “Application Programming Interface” is a programming interface provided by a software system. It enables developers to use or extend the available functionality of the software system by own software programs.
Arithmetic Mean (also “Mean Value” or “Average”)
The arithmetic mean can be calculated by dividing the sum of all values by the number of values. When analyzing load test results, e.g. response time, the arithmetic mean alone is not sufficient as single extremely deviant values can influence the arithmetic mean too much. For five responses an arithmetic mean of 5s could be the result of all request answered in 5s. But it could also be the result of one response time of 21s and four responses with a response time of 1s each. Therefore also other metrics like minimum value, maximum value, standard deviation (root mean square) and median should be considered.
P: A bottleneck is a hardware or software component that is so busy or so inefficient that it is mainly responsible for the systems limitation in performance. Optimizing this component or providing this component with more power will increase the performance of the complete system. No bottleneck exists if all components are evenly busy and have a share in response behavior of the complete system.
Capture & Replay Tool (also called “Capture & Playback Tool”)
This is a category of testing that is recording interactions with the system under test manually performed by a tester. The recorded interactions can be replayed automatically later on. Capture & Replay alone is not sufficient in most cases and therefore existing as an additional tool in a broad testing tool or it can be extended by further editor functionality. Recorded test cases often have to be extended manually to get robust and reusable test cases.
Main problems with pure recorded test scripts are:
- Lack of dynamic - e.g. in an online shop the same product pages a accessed again and again instead of selecting from a list randomly or crawling dynamically.
- Low robustness against changing test data - e.g. missing ability to react to changing number of results for search phrases; for zero results you might see the search form again, for just one result you might see the result page immediately, for several results a list to select from appears.
- Low robustness against changes in the system under test itself
- Collection of many small, independent test cases instead of one structured test suite. Missing reusability and difficult maintenance
- Often insufficient validation of results
The XLT Script Developer was such an advanced Capture & Replay Tool, featuring a broad range of additional functionality, but it was discontinued when Mozilla removed the XUL-application API from Firefox.
A client is a process in the client-server-model that sends requests to another process (server) to make demand on provided services. Clients of a web server are browsers, search engine bots or, when running an XLT load test, the virtual users with the headless browsers.
Concurrent Users are several real or simulated users, which are using the system at the same time during a load test. The terms “Parallel Sessions” or “Parallel visits” are also used.
For web applications the number of concurrent users calculates as follows: Each user sends single requests, gets a response, and then waits for a certain time (think time) until he sends the next request. Nevertheless the user counts as a concurrent user during this time. The parallelism encloses all users that started a visit on the web site (first request) and did not finished it yet, thus will send more requests. Thereby it is not relevant whether a user’s request is still being processed or the last response has already arrived and no new request has been sent yet.
If the real user scenarios contain a long think time between the requests then possibly a simulation with less virtual users but shorter think time is sufficient. The aim in this case is to simulate the same count of parallel requests. Such a scenario is not a full equivalent to a higher number of users with longer think time but depending on the aim of the test it can be sufficient in most cases.
Constant Arrival Rate
Constant Arrival Rate is a kind of simulated load where the virtual users start new visits with a constant rate, independent from the current operating state of the system under test. The number of parallel visits depends on the response time behavior of the system under test. Even if the system responds very slow or not at all, new visitors will arrive hence sending requests to the system. In theory the number of parallel visits could rise indefinitely.
If the tested system can process the load the system will engage to a certain level of parallel visits. This state is reached if the number of finished visits per time is the same then the number of requests for new visits arriving at the server.
If the tested system can not process the arriving requests fast enough the number of parallel users increases automatically. This type of simulated load is close to real systems behavior because the number of new arriving visitors is also not depending on the actual state or speed of the server in real life.
When using XLT a cluster wide constant arrival rate is simulated if the
com.xceptance.xlt.loadtests.default.arrivalRate is set and no
iteration count is configured by the property
com.xceptance.xlt.loadtests.default.iterations. This property defines
the number of new visits per hour. The maximum number of parallel
virtual visitors is limited by the mandatory property
com.xceptance.xlt.loadtests.default.users, also for the simulation of
a constant arrival rate.
A Cookie can be transient or persistent. Transient cookies are stored in the main memory of the browser process so that they are only existing during the runtime of the web browser. In contrast persistent cookies are stored to the file system by the web browser. The web server that creates the cookie can set a desired lifetime for the cookie. So the cookie is still available even after restarting the browser or the client computer.
XLT test cases not run by a mastercontroller but by any JUnit test runner directly as a JUnit test automatically run in development mode. E.g. when running XLT test cases in Eclipse or as ANT task. When using the development mode additional property files are read. So it is possible to redefine settings for local test runs during development without changing the settings of the load test mode.
DOM (Document Object Model)
Efficiency is a quality measure regarding ISO 9126. It describes the performance of a software system in relation to the used resources. It is divided into sub-characteristics: time behavior and resource behavior. Time behavior describes the response and processing time as well as the throughput. Resource behavior describes the amount and the duration of neded resources.
An end-to-end test is testing a complete system from the point of view of an end-user or a user making demands on the service. Such a test includes all components involved to provide the service.
To perform an automated end-to-end test or a end-to-end monitoring a test software simulates the behavior of the users. A request passes through the system from the client, through all intermediate layers to all involved servers and back end systems until the response finally arrives at the client again (end-to-end). There the response can be verified, e.g. regarding availability, response time and functional correctness.
Endurance tests are load tests running continuously over a longer period. Purpose is to test the long-term stability, the resources requirements (e.g. memory leaks) and the response time behavior of the system under test. The test period is often 12 hours up to several days. Often a load of 30% below the maximum load determined in a stress test is simulated during an endurance test. Endurance tests are also called stability tests.
A failover test checks high availability capabilities of an application. To do so single redundant designed system components are forced to break down and re-start again during a load test. The correct system reaction during this scenario (failover) can then be checked.
“Favicon” is an abbreviation for “Favorite Icon”. It is a graphic in the
windows icon format which can be displayed in the web browsers address
bar or next to a browser bookmark (also called favorite). The favicon
file has to be available in the root directory of the domain:
http://www.mysite.com/favicon.ico. It is also possible to define an
URI to the favicon file in the header of a html file:
<link rel="shortcut icon" type="image/x-icon" href="/favicon.ico">
Many browsers automatically start a request to the favicon when the user bookmarks a link from that domain. If the server cannot find the file it sends an http code 404. Initially the favicon was introduced by Microsoft with one of the earlier versions of Internet Explorer.
See Local Shared Object.
When creating XLT test cases sometimes you may want to reuse blocks of
code that contain more than a single action.
For this purpose you can create it’s own class with one method that combines a sequence of several XLT actions as a so called “flow”. Different test cases can call this method to reuse the flow. This can be implemented if needed, but there is no explicit support available or necessary in the XLT framework.
The term is often used for the automatic request and loading of images
and other multimedia resources by a web browser. A HTML files contains
no image files. To display images on a web page the browser is parsing
the loaded HTML file and starts an automatic request for the embedded
image URLs to load and display the images. The image loading can also be
defined by CSS rules. In XLT the image loading can be configured with
com.xceptance.xlt.css.download.images. Sometimes the term “image
See also Static Content Loading.
Initial Delay (XLT)
The initial delay is a time period in the beginning of a load test where neither load is generated nor measurements are taken. It is a pure wait time before any activities. This can be useful for fully automatically started complex test runs, e.g. to make sure a also automatically started system under test has reached stable operating conditions.
The term instrumentation is used for adding code to the tested application for gathering runtime information about the system under test. Examples are information about passed through code branches (code coverage) or the frequency and duration of method calls (profiling).
JUnit is a framework for automated unit tests of Java applications. It was originally developed by Kent Beck and Erich Gamma. Test cases written in Java following the JUnit conventions can be executed by any JUnit test runner.
A Load Profile is a description containing all relevant load parameters like number and type of virtual users, the simulation of constant arrival rate or stead load, a ramp-up period, think time etc.
When performing a load test the future use of the tested system is simulated and assessed considering a certain number of users and transactions. This is done for two reasons. The first is to recover functional failures appearing only under parallel and intense usage of the system. The second is to measure time behavior and resource behavior under a certain load. Thereby values are determined like response time, main memory usage or processing performance.
Load tests can focus on different system aspects. Different load profiles and test scenarios are necessary for the different aspects, often named with an special term like “stress test”, “endurance test” or “scalability test”. Load test is the generic term for all of these tests.
Load Test Mode (XLT)
XLT works in the load test mode if test cases are executed by the XLT master controller and the XLT agents. XLT auto-detects the mode depending on the used test driver. The load test mode is mainly used for load test, hence the name. For test cases running in load test mode, the additional property files for development mode will be ignored.
Local Shared Object
In the web development environment a “Local Shared Object” is a file that is stored to a users local computer with the help of Adobe Flash player when the user visits a web page with the relevant Flash content. Therefore these files are also called “Flash cookies”.
Local shared objects are used for the same purpose as HTTP cookies and follow the same rules. They can only be read by the web site that initiated to save the local shared object. In contrast to HTTP-cookies, they are browser independent. Local shared objects saved in one web browser can be read by the Flash player in another web browser and then be sent to the web server. Furthermore, local shared objects can hold a lot more data and they do not have a expiration date. Saving local shared objects can be prevented by most of the web browsers at the moment.
Master Controller / Mastercontroller (XLT)
The mastercontroller controls the entire load test. It deploys the test suite to all agentcontrollers, calculates the load distribution, and starts and ends the test. At the end, the mastercontroller is gathering the test results and stores them on the file system.
To be better highlight that a
mastercontroller is a shipped XLT component, we prefer the spelling mastercontroller over master controller but you might see both spellings used.
Measurement Period (XLT)
Time period for taking measurements (com.xceptance.xlt.loadtests.default.measurementPeriod). The measurement period begins after the warm-up period.
The median is the value in the middle of a distribution of values. The median halves the set of values as follows: Not more than half of the values are greater than the median and not more than half of the values are less than the median. To determine the median, all values will be sorted. For a uneven number of values the median is the value in the middle of the list. For a even number of value the median is the mean value of the two values in the middle of the list. In comparison to the arithmetic mean the median is a lot more robust against single deviating values. Example: The median of five response times < 1s, 1s, 1s, 1s, 21s > is 1s. The arithmetic mean would be 5s.
A memory leak is a software fault that does not free (de-allocate) main memory areas after using it. If such a fault takes affect repeatedly during runtime, the memory demand for this application will grow until there is not enough free memory available for further program execution, ultimately leading to a software failure.
A metric is a measure to cover quantifiable properties. Every metric is based on a definite, reproducible measurement instruction. Examples for metrics in the load test environment are the reached number of requests per second, the mean response time in seconds or the CPU usage in %.
See Page View.
Parallel requests are all requests sent but not completely responded. The number of parallel requests is often smaller than the number of concurrent users. It can also be bigger because one request sent by a user to load a web page can trigger several parallel requests automatically to load static content like images.
The term “performance test” is not always used in a uniform manner. It often describes a test to check the compliance of given load requirements by simulating the defined load and comparing the systems behavior with the requirements regarding response time, throughput etc.
The term performance test is also used if the test should find bottlenecks without overloading the system. For that purpose detailed runtime monitoring information about all hardware and software components will be gathered and assessed.
Sometimes the term is used interchangeably to the generic term load test.
A percentile defines how many numbers (measurements) fall below or above a certain percentage of the data and the maximum number of this data is taken as the resulting value. For instance a P90 defines that out of 90% of all measurements (sorted ascending), the highest value for the first 90% of values is used as the result.
For instance, when we have the measurements:
50, 60, 65, 100, 500, 15, 60, 10, 1000, 100
we sort them first:
10, 15, 50, 60, 60, 65, 100, 100, 500, 1000
and read the value at 90%, in this case the 9th value. If we have 100 values, it would be the 90th value, just to make it simpler. In this case, our P90 is 500. Our P50 would be 60 and so forth. The mean (average) is 196.
As you can see, the P90 value more precisely shows that there is a problem with 10% of the measurements or requests or sessions, whatever value you are looking at.
Property files are text files in a defined format which are used to save settings in the Java environment. These files contain one property per line, each with a name and a value.
QA is an abbreviation for “Quality Assurance”. It covers all actions and methods to achieve and control a desired quality. Sometimes the involved organizational units are called QA, e.g. test teams.
Ramp-Up Period (XLT)
The time period to automatically ramp up the number of virtual users to
com.xceptance.xlt.loadtests.default.rampUpPeriod). It can be
used to simulate a slowly increasing load, e.g. to determine the maximum
processable load. This setting is independent from warm-up period,
measurement period and shut-down period. The ramp-up period starts after
the initial delay.
Ramp-Up Step Size (XLT)
The number of virtual users by which the load is automatically increased
with every step (
The time between the steps is calculated automatically by XLT. With the
last step just before the ramp-up period times out the configured number
of virtual users is reached.
A regression test is repeatedly execution of a test case regarding requirements against a already tested software to discover unwanted side effects. According to the requirements only a part of the test cases or all previously executed test cases can be repeated.
A Request is an inquiry from a client to a server mostly using the HTTP or HTTPS protocol. Also see Hit.
The term Saturation is used if a resource or a system is fully used to capacity or completely busy. If this state is reached an increased load will not result in a higher throughput.
Scalability describes how effectively an application can process additional workload by using additional resources, without the need of further changes on the application.
Ideally, the processable workload of an scalable application is linearly increasing with the amount of available resources. E.g., such an application could respond twice the number of requests per time by doubling the available resources. This is called “linear scalability”. Necessary prerequisite for linear scalability is that all resources can work to a high capacity at the same time.
In real life synchronization, communication and administration overhead leads to the fact that added resources can not be used by the application completely. This results in “sub-linear scalability” where doubling the resources increases the processable workload by less than 100%.
A scalability test evaluates how a systems response behavior and consumption of resources changes with increasing load. Ideally, the response time increases linearly with the load. In this operating range we talk about a “scaling system”. In real world systems this ratio gets worse with increasing load. From a certain point the response time is disproportionately increasing with the increasing load. This is the limit of scalability. The scalability test evaluates how much we can increase the load for a given system until we reach this point.
Furthermore the scalability test checks whether and how far this limit can be pushed up by adding additional hardware and software resources. As long as the maximum processable workload is approximately increasing linearly with the available components we talk about a system “scaling over the hardware and software.
See Test scenario.
A server is a system providing services that clients can make use of by a defined interface. For web applications that could be a web server or application server taking requests from clients and delivering HTML pages and other content as response.
Service Level Agreement (SLA)
A “Service Level Agreement” is a binding agreement between a service provider (contractor) and a service user (principal) regarding certain performance characteristics, e.g. availability or maximum response time.
A session designates all activities related to one user during a visit and also their technical representation on server side. For dynamic web applications beyond the simple display of static web content all requests sent to server during a visit have to be assigned to that certain visit by the server. Because of using stateless protocols and a lack of a unique ID available with every request a user in the web can only be recognized with the help of additional methods.
Therefore, the server generates a unique session ID for every single user when the first request from that user arrives at the server. From this point the session ID will be transferred together with all requests and responses between client and server. Depending on the application the server saves internal status together with the session ID, e.g. the basket of a online shop, login information or the selected language for multi language sites.
The end of a session often is not clearly determinable because it is not known whether the user will send further requests. Therefore sessions will often be finished on server side after a certain time if no further request has arrived (session timeout). In this case the session data will be deleted or marked on server side. If a session ID has been generated but the user has not performed an action that uniquely identifies him (Login) this is called a anonymous session.
Session ID (SID)
A session ID is a unique identifier for visitors on a web site. It is used to identify several requests related to one user. A session ID will be generated when the the first request of the user is is being processed by the server and then it will be send back to client together with the response. From this point the session ID will be transferred together with all requests and responses between client and server. The session ID can be transferred by a cookie in the HTTP header, in the URL or as a hidden form parameter in the POST-data.
A session ID must be unique and very hard to guess. Furthermore additional mechanisms should be used to prevent taking over a foreign session by attackers.
Shutdown Period (XLT)
The shutdown period is the time period at the end of a load test where
the virtual users finish the already started transactions but no
measurements will be made
com.xceptance.xlt.loadtests.default.shutdownPeriod). The shutdown
period is used to allow the user transactions quitting cleanly after the
measurements have been made. Transactions still running after the
shutdown period has timed out will be stopped abruptly.
See Virtual User.
The aim of a sizing test is to find the needed hardware and software resources for certain load requirements. For this purpose the limits of several different hardware/software configurations will be determined so that later, you’ll have an approximation for the prospective hardware and software resources for given load requirements.
Standard Deviation is a measure for the deviation of all values. It gives an information about the mean distance of the values from the arithmetic mean. It calculates as the square root of the variance of the data.
When assessing the deviation of measured values the standard deviation is always to be considered in relation to the arithmetic mean. A comparatively high standard deviation shows a high variation of the values.
The values are called “normally distributed” if the distance from the arithmetic mean is not exceeding the standard deviation for 68% of the measured values and furthermore if the distance from the arithmetic mean is not exceeding double the standard deviation for 95% of the measured values.
If the measurement values for a response time with a arithmetic mean of 5s and a standard deviation of 2 are normally distributed, then 68% of all response times are in the range of 3s..7s, and 95% of all response times are in the range of 1s..9s.
Static Content Loading
A test run without loading static content is used to test just the application logic and the generation of dynamic web pages, instead of testing the delivery of static contents which is performed by other system components in most cases. This saves resources on server side, in the network, and on client side and therefore with less resources a higher number of virtual users can be simulated. Thus, the tested core components can be stressed with a higher load.
XLT provides several properties to control loading static content (see
com.xceptance.xlt.loadStaticContent). Note: If the property
true all virtual users
com.xceptance.xlt.loadStaticContent is set to false.
For this kind of simulated load a constant, exactly defined number of virtual visitors is used. Each virtual user runs several transactions (test cases) in succession. A virtual user can start a new transaction only if the the previous transaction is finished. In this case even a very long response time does not lead to a higher number of parallel visitors.
This mode is not very close to real life but is advantageous for
controlled measurements because it minimizes high variability and
uncontrolled side effects. XLT simulates a steady load if the property
com.xceptance.xlt.loadtests.default.arrivalRate is not set.
A stress test is increasing the simulated load beyond the limits of the expected normal operation until functional failures appear or until the response time of the system under test is exceeding defined limits.
For this purpose often a continuously increasing number of users is simulated. Stress tests are used to evaluate the systems behavior in overload situations and after it, to determine the maximum acceptable load or to find performance flaws (bottlenecks).
See Local Shared Object.
System Under Test (SUT)
The hardware and software of the tested system, e.g. the demo application “Posters”, distributed together with XLT, installed on a certain computer.
Test Case (XLT)
Using XLT, a test case is a piece of code that implements a user transaction and that can be run with any JUnit test runner (Java test case classes).
A test scenario is a sequence of steps in the application under test performed by real users or virtual users. For example these are use cases like browsing or the registration of an new user in a online shop.
Sometimes the term is also used in a more broadly manner, e.g. to describe a complete load profile with a mix of several virtual user types.
Test Suite (XLT)
A collection of test cases belonging together and to be performed at once.
Think time is the time span between arrival of a response sent from server to client and the next request sent to the server. From a server’s point of view, the think time is an interrupt of the user’s action. A real user for example needs time to read text on the requested web page, to click a link or to fill a form. When using XLT, separate values for think time between actions and think time between transactions can be configured. Furthermore the think times can vary randomly using XLT.
Throughput is a measure of the amount of requests or data that can be processed or transferred in a defined time range. When running load tests this is, for example, the number of processed requests per time range, or the amount of data per time range, transferred from the server to the virtual users.
Execution of a certain test case that models a test scenario.
See Test Scenario.
A virtual user is a simulation of a real user (and his used the application) by a test tool. The virtual users should realistically model all characteristics relevant for the test case.
During a load test a number of virtual users run parallel and execute test cases independently.
A visit is a sequence of a users page views on a web site temporally interrelated.
A visit to an online shop is comparable to visit to a real world retail shop. The user enters the shop (first request), possibly searches some products, views single products in detail, puts them into shopping cart and eventually buys these products.
All requests send to the server during one visit have to be assigned to that certain visit by the server. Therefore the server generates a unique session ID for every single user when the first request from that user arrives at the server. From this point the session ID will be transferred together with all requests and responses between client and server. Depending on the application the server saves internal status together with the session ID, e.g. the basket of a online shop, login information or the selected language for multi-language sites.
The term “visit” is often used for statistical evaluations, in contrast the term “session” is often used in the technical environment. Both terms are also often used synonymous.
From a technical point of view there is no defined end of a visit because it can not be predicted whether the user will send a further request. Therefore session will often be finished on server side after a certain time if no further request has arrived (session timeout). In this case the session data will be deleted or marked on server side. Depending on the web service this timeout is often in a range of 30min to 24hours. The next request of the same user after the session has timed out creates a new session (a new session ID) and will be assigned to a new visit.
This is why the last visit must be some time in the past to make another visit an actually new visit.
A visit has at least one page view, but has several in most cases. Most important measures are the duration of a visit, the number of page views and the time lag between the page views (think time).
Warm-Up Period (XLT)
Time period where load is already generated but no measurements are
com.xceptance.xlt.loadtests.default.warmUpPeriod). This can
be used to configure a settling time where response time and other
measured values are not yet representative and hence should not be
considered in the load test results. The warm-up period starts directly
after the initial delay.
A web site is one complete, interrelated web presence. Generally, a web site designates all web pages and available services of a company or organization.
WebDriver API is a tool for automated test of web applications. Either real web browsers can be controlled or web users can be simulated. Google WebDriver features an easy and efficient usable API. For a quick creation of simple test cases also the XLT framework provides the usage of the WebDriver API.
Therefore external test cases using the WebDriver API can generally also run in the XLT framework. The other way round this is not true in all cases because in the XLT framework WebDriver API has been expanded by the concept of action names. Test cases programmed in XLT using the WebDriver API must possibly be adapted to run with another WebDriver-compatible test driver.
XPath is an abbreviation for “XML Path Language”. XPath is a query language to address elements of a XML document. In XLT XPath can be used to address elements of a loaded HTML page.
The file extension “XPI” (Cross-Platform Install) has been developed by the Mozilla foundation. XPI files are zip-compressed installation files. They contain installation scripts and other added files.
The “Extensible Stylesheet Language” is a set of transformation languages for the purpose of transformation or presentation of XML documents. Members of this language family are XSL-FO (XSL Formatting Objects) for formatting instructions to export XML documents, XSLT (XSL Transformations) for transformation of XML documents in other XML documents as well XPath to address elements inside a XML document.