There are many instances where you get reports indicating higher page response times in the IBM Rational Performance Tester (RPT) generated reports. You tend to correlate these metrics against the usual time behavior noticed when accessing the application under test outside RPT. It could be a case as well where even if you add all the page elements listed in report, the response time is much less compared to overall response time for the transaction.
So let's see how RPT as a load generation tool computes response time metrics.
Generally, the response time is measured from the time the first byte leaves the client machine and until the time that the last byte is returned.
The page is considered to begin when the first action (typically a request) associated with the page starts execution. If this request needs to make a connection, the page response time will include the connection time. The page ends when the last action (usually a response) of the page completes. RPT would adjust the page response time to remove the time spent in client delays. In RPT 8.2 this changed after determining it was more representative of the real-world response time to leave the client delays in the page response time. Thus, in RPT 8.2 and later, the page response time does include client delays.
However the Think Time value set in the schedule is not included in the page response times displayed in the Performance report. It should be noted that there are potential "delays" in the script that could increase page response times. Think times are associated with pages and are intended to represent human pauses during recording - these don't impact page response times. For individual page elements (requests), there may also be delays which are intended to represent client (browser) delays (processing time for example). These will be reflected in page response times
In addition to this, connection times are included in page response times. Connections times are, by design, not included in page element response times since a particular page element may make a connection sometimes but not make a connection other times. Here's a brief explanation on why the page response times are not equivalent to the sum of its requests.
Questions could arise such as: How does RPT calculate the Page level response time as compared to Request level response time for a given page?
In case the first request used a new TCP connection, it calculates the difference between the timestamp when the connection for the first request in the page was made and the timestamp of when the last byte was received for the last request in the page. In case an existing connection was reused, the timestamp of when the first request was sent is used. All these timestamps are available in the test log if full logging is on. If a existing TCP connection is re-used, the value "Time to Connect" in the test log will be 0 .