Performance measurements
To ensure correctness and sufficient precision of the measurements, the testing procedure should meet the following requirements:
- The CPU load of the computer running the tests should not exceed 90% (during test run).
- If the amount of virtual clients is greater than 1, the start and end points of the measurement period should be located within the time period when ALL virtual clients are active (see the description of delaystart, maxrequests).
Recommended settings for the [General] section:
[General]
delaystart=0
Provides simultaneous start of virtual clients (see description of delaystart)
maxrequests=50000
The maxrequests value should not exceed the product (clients x requests). Required to eliminate the “final steps” (see the description of maxrequests)
nowait=0
Do not use the timeout check. Allows to provide significant server load using a single workstation, thanks to saved resources. Consequently, can lead to hang-ups of virtual clients in case of a big amount of stuck requests (which would be normally closed using the timeout check)..
log=0
Logging is off - to reduce resource consumption of the workstation running the test.
nodetails=1
Turns off the printing of detailed statistics on separate requests for all virtual clients. Useful if measuring only the value of Requests per second.
summarytimes=1
Turns on printing of the summary statistics
Example: a test script for measuring the performance of loading the main page of a Web server:
[General]
clients=50
requests=10000
server=myserver
keepalive=1
delaystart=0
maxrequests=50000
nowait=0
log=0
nodetails=1
summarytimes=1
[TestSeq]
Req1
[Req1]
$uri=/