Response Probe writes its results to a text output file. The default filename is the same as the Process script file name, but with the .out suffix. However, you can enter a different name as part of the command to run Response Probe.
The file begins with a header that includes copyright and version information, the date and time the test was run. The data lines appear below the header. One line of data is printed for each thread in the Response Probe test.
Response Probe times cycles, each of which consists of one think state, file access state, and compute state for the workload. The output values represent the time of workload cycles only. Response Probe computes and subtracts its own overhead from the values before displaying them.
The values in the output files are actual values, not necessarily those specified in the Thread Description file. Response Probe calculates values based on the means and standard deviations in the Thread Description file. When it calculates an uninterpretable value, such as a negative page number, it folds (recalculates) the number into one that makes sense. For more information, see "How Response Probe Works," later in this chapter.
The following figure is a sample output file. In an real file, the data line columns appear in a single row.
Multi-Processor Response Probe.
Copyright 1990-1993 Microsoft Corporation.
Version 2.1 (93.12.21)
Wed Mar 15 05:45:14 1997
Script File : 512.scr
Trial Time : 300 seconds
Stable interval : [50%..100%] of Trial Time == [150..300] (150 seconds)
Relative Processor Speed: 4.31
(All times are in milliseconds) | | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
| | | | | | |
Item | Definition |
Script File | The filename of the process script file (*.scr). The process file includes the name of thread definition file, and the thread definition file includes the name of the thread description file, so all Response Probe script files used in this test can be traced from this single entry. |
Trial Time | This is the time limit you set for the test on the command line. The actual test time is shown in Total Time in the output table, below. |
Stable Interval | The time that elapsed before Response Probe started measuring. This is half of the value of the TrialTime parameter in the run command. Response Probe assumes that, by that time, all threads have launched and are processing, and increasing the time limit on the trial does not significantly change thread response times. |
Relative Processor Speed | The number of times the compute loop must be executed to consume one millisecond of time on the processor. This is calibrated when Response Probe starts and is used to run the compute state for the time you chose. |
PID | Process ID. Windows NT assigns an identifier to all processes. The IDs are assigned in the order that the processes start and don't imply any priority. This field is for identification only. |
TID | Thread ID. Windows NT assigns an identifier to all threads in a process. The IDs are assigned in the order that the threads execute and don't imply any priority. This field is for identification only. |
File mode | A combination of FILEATTRIBUTE and FILEACCESSMODE. Reports whether file access was random or sequential, and whether access was buffered, unbuffered, or mapped. This field should match the input in the Thread Definition file parameters. |
Rec size | Size of the each read/write as specified in the RECORDSIZE parameter, in bytes. This field should match the input in the Thread Definition file RECORDSIZE parameter. |
Total time | The actual elapsed time for the test. Compare this value to Trial Time in the output header. |
Resp time | Total elapsed time for all test cycles during the measured part of the test, that is, half of the trial time. Response Probe calculates its own time and subtracts it from the data. |
Resp cnt | Number of cycles completed during the measured part of the test. |
Mean | Average elapsed time of a cycle. This value should be the same in repeated trials with the same input. If it is not, lengthen the trial time of the test. |
Sdev | The standard deviation of elapsed time of a cycle. |
Min | The elapsed time of the quickest cycle. |
Max | The elapsed time of the slowest cycle. |
Think mean | Average elapsed time of the think states in the measured cycles. |
Think Sdev | Standard deviation of the elapsed time of the think states in the measured cycles. |
Reads mean | Average elapsed time of the file access states (reading and/or writing) in the measured cycles. |
Reads Sdev | Standard deviation of the elapsed time of the file access states (reading and/or writing) in the measured cycles. |
CPU mean | Average elapsed time of the compute states of the measured cycles. |
CPU Sdev | Standard deviation of the elapsed time of the compute states of measured cycles. |
DataPg mean | Average elapsed time of writes to Response Probe's internal data pages. |
DataPg Sdev | Standard deviation of the elapsed time of writes to Response Probe's internal data pages. |
Func mean | Average elapsed time of reads from Response Probe's 1000-function simulated codepage file. |
Func Sdev | Standard deviation of the elapsed time of reads from Response Probe's 1000-function simulated codepage file. |