Information Technology Reference
In-Depth Information
logs. When such logs are not available, rough failure data can be extracted from
access logs.
Assessing the operational reliability for web software
: When used with fail-
ure data to estimate failure rate or reliability, all four workload measures used
in this section produced more consistent and stable reliability estimates than
using daily errors alone. They offer reliability assessments from different per-
spectives, and each may be suitable for specific situations. For example, byte
count might be more suitable for traffic measurement and related reliability in-
terpretations; hit count might be more meaningful to web users as they browse
individual pages; while number of users or sessions might be more suitable for
high-level web site reliability characterization.
Assessing the potential for reliability improvement under effective testing
:Also
demonstrated in both case studies is the significant potential for reliability im-
provement if defect can be fixed using our testing strategy described in Sec-
tion
5
.For
www.seas.smu.edu
,
the failure rate could be reduced to 42.1-25.2%
of the initial failure rate through such testing equivalent to 26 days of operation.
Similarly for
www.kde.org
,
the failure rate could be reduced to about 11-13%
of the initial failure rate through testing equivalent to 22 days of operation; or,
equivalently, almost all the original problems could have been fixed. These re-
sults provide external validation for our integrated testing strategy described in
Section
5
.
Some generalization beyond our two case studies
: Many of the results we ob-
tained and patterns we observed concerning workload measurements for the
SMU/SEA and the KDE web sites are remarkably similar to that for other In-
ternet traffic
[1,12,21,39]
, which indicates that web traffic characteristics have
remained fairly stable for almost a decade. Although re-confirming these exist-
ing results and patterns is not our intention or our focus, this confirmation lends
further validity to our primary purpose of using these measurements as part of
the data input to evaluate web software reliability.
7.
Conclusions and Perspectives
To summarize, a collection of appropriate testing techniques can be selected,
adapted, and integrated to help us perform effective web testing and to ensure web
reliability. As we demonstrated in this chapter through our case studies using the web
sites for the School of Engineering and Applied Science, Southern Methodist Uni-
versity (SMU/SEAS) and for the large-scale open source KDE project, hierarchical
testing of web-based applications is both viable and effective:
