Output appears in the following format in a text file named as indicated previously in "Performing Response Probe Experiments." For each thread in the experiment, one data line is printed.
|
Multi-Processor Response Probe.
Copyright 1990-1993 Microsoft Corporation.
Version 2.0 (93.06.24)
Wed Jun 24 15:36:02 1993
Script File : pb01_01a.scr
Trial Time : 100 seconds
Stable interval : [50%..100%] of Trial Time == [50..100] (50 seconds)
Relative Processor Speed: 11.37
(All times are in milliseconds)
File Rec Total Resp Resp
PID TID Mode Size Time Time Count Mean SDev Min Max
----- ----- ------ ---- ----- ----- ----- ---- ---- ---- ----
72 71 S U 4096 100049 46235 941 49 0 49 50
Think Think Reads Reads CPU CPU DataPg DataPg CodePg CodePg
Mean SDev Mean SDev Mean SDev Mean SDev Mean SDev
----- ----- ----- ----- ----- ----- ------- ------- ------- -------
0 0 0 0 50 0 1 0 1 0(In the actual output file, the columns from "PID" to "CodePg SDev" all appear in one wide row.)
The really interesting number here is the "Mean" response time to the workload that you have devised. This is in the eighth column from the left. In addition to all the Performance Monitor data you may collect when you run your experiment, this number can be revealing. Keep an eye on it.
You want to be sure your experiment is long enough that the "Mean" response time for the action you specified is repeatable. Keep increasing the length of your experiment until this is true. The larger the number of threads and the more file or paging activity you generate, the longer your experiments will have to be.
"PID" is the Process ID of the thread, and "TID" is the thread ID. "Total Time" is the experimentally observed total time for this thread. "Resp Time" is the time during which the thread actually observed its own response time. It should be the last half of the total time. "Resp Count" is the number of complete cycles observed in the response time computation. "Mean" is the average response time to the Resp Count cycles. "Sdev," "Min," and "Max" are the standard deviation, minimum, and maximum of the response time observed in the Resp Cycles. The remaining columns specify some of the input parameters to the experiment, in case you lose your .SCT file.
Response Probe should help you get a good idea of how your equipment can handle workloads of various types, as well as how changes to your equipment have affected its capacity. What a tool!