Web Testing for Reliability Improvement - Advances in Computers - page 209

6. Web Workload Characterization and Reliability

Analyses

For software systems under normal operation or in testing, the execution results
can be analyzed to assess test effectiveness and product reliability. In general, the
failure information alone is not adequate to characterize and measure the reliability
of a software system, unless there is a constant workload [27,34] . Due to the vastly
uneven web traffic observed in previous studies [1,39] , we need to measure both the
web failures and related workload for reliability analyses. We next adapt existing
techniques that characterize workload for traditional software systems and analyze
their reliability to the web environment [52] . These measurement and analysis results
also provide external validation or effectiveness assessment for our integrated testing
strategy described in the previous section.

6 . 1 D e fi n i n g W o r k l o a d M e a s u r e s f o r R e l i a b i l i t y A s s e s s m e n t

The characteristics of the web environment discussed in Section 3 require us to
measure actual web workload to ensure its satisfactory reliability instead of indis-
criminately using generic measures suitable for traditional computation-intensive
workload. The user focus and substantial amount of idle time during browsing
sessions make any variation of execution time [34] unsuitable for web workload
measurement. Similarly, the dominance of non-computational tasks also makes com-
putational task oriented transactions [51] unsuitable for web workload measurement.
Instead, other workload measures may be more suitable for characterizing workload
at web sites.
From the perspective of web service providers, the usage time for web applica-
tions is the actual time spent by every user at the local web site. However, the exact
time is difficult to obtain and may involve prohibitive cost or overhead associated
with monitoring and recording dynamic behavior by individual web users [39] .One
additional complication is the situation where a user opens a web page and contin-
ues with other tasks unrelated to the page just accessed. In this situation, the large
gap between successive hits is not a reflection of the actual web usage time by this
user. To approximate the usage time, we can consider the following workload mea-
sures [52] :

Number of hits, or hit count. The most obvious workload measure is to count
the number of hits, because (1) each hit represents a specific activity associated
with web usage, and (2) each entry in an access log corresponds to a single hit,
thus it can be extracted easily.