Information Technology Reference
In-Depth Information
The purchase step transaction group is behaving well (not slowing down much)
throughout the workload. The log on/log off transaction group behaved well until the
workload reached 400 users. At that point, the log on/log off performance jumped
to the requirement maximum. The performance tester should use this information to
alert the developer that the log on performance becomes marginal when the purchase
step workload exceeds 400 active transactions.
Figure 9.8 shows the results of ramping up the log on/log off along with launch-
ing purchase steps and catalog browser transactions to peak workload.
Figure 9.8
Round trip performance for peak logon + purchase + catalog browse workload mix
9.5.4 Saturday Workload Showstoppers
Clearly, there is some kind of resource confl ict between the purchase steps and the
catalog browse. The performance requirement max for the catalog browse was ex-
ceeded when the ramp-up achieved only 100 active users. These results are suffi -
ciently severe as to halt further performance testing of the Saturday workload until
the development team diagnoses and implements some performance solutions. This
kind of testing discovery is referred to as a “showstopper” from the testing perspec-
tive. Furthermore, because we chose the lower volume workload peak to test fi rst, we
know that there is no benefi t to starting the weekday workload testing that demands
an even higher volume of the same poorly performing transactions.
While the performance testers await development solutions to these response
time problems, other types of testing such as functional and structural testing can
continue. Notice that testers doing the other types of testing will be affected when
the developers do fi nd solutions to the response time problems. Specifi cally, the de-
velopment solutions to the performance problems will need to be retested (regression
testing) to verify that the performance solutions affect only the performance aspect
of the software and not inadvertently impair any functional aspects of the software.
Regression testing here will add time and effort to the test schedule that may not
have been anticipated in the original test execution schedule.
Search WWH ::




Custom Search