A Case Study in Integrated Quality Assurance for
Performance Management Systems
Liam Peyton, Bo Zhan and Bernard Stepien
School of Information Technology and Engineering,
University of Ottawa, 800 King Edward Avenue, Ottawa, ON, Canada
Abstract. On-line enterprise applications that are used by thousands of geo-
graphically dispersed users present special challenges for quality assurance. A
case study of a hospital performance management system is used to investigate
the specific issues of architectural complexity, dynamic change, and access to
sensitive data in test environments. A quality assurance framework is proposed
that integrates with and leverages the performance management system itself.
As well, a data generation tool suitable for the requirements of testing perfor-
mance management systems has been built that addresses limitations in similar
commercially available tools.
1 Introduction
Increasingly, large organizations are deploying complex on-line enterprise applica-
tions that are used to run and manage critical business processes [5]. These applica-
tions often run on a network of centralized servers in a service oriented architecture
(SOA) [12], and are accessed by thousands of geographically dispersed users via
browser-based interfaces. In this case study, we examine some of the specific re-
quirements and challenges faced in providing a quality assurance framework for a
performance management and reporting system deployed at a large teaching hospital.
This hospital has thousands of health care workers and maintains a data warehouse
containing hundreds of millions of records related to patient treatment. The perfor-
mance management system in the case study uses typical off the shelf components
that are used in other hospitals, and which are used quite regularly in deployments for
the biggest companies in the world (e.g. members of Forbes magazines "Global
2000").
Business performance management (BPM) enables an organization to understand
the status of business processes across business and IT, in context against goals and
trends, in order to improve business operations [6]. BPM is often used to enable spe-
cific management initiatives like balanced scorecard, Total Quality Management, Six
Sigma, and Sarbanes-Oxley. Enterprise performance management systems that sup-
port BPM provide services to manage and report on data collected from across the
organization's business operations into enterprise-wide data warehouses and data
marts.
Peyton L., Zhan B. and Stepien B. (2008).
A Case Study in Integrated Quality Assurance for Performance Management Systems.
In Proceedings of the 6th International Workshop on Modelling, Simulation, Verification and Validation of Enterprise Information Systems, pages
129-138
DOI: 10.5220/0001743701290138
Copyright
c
SciTePress
A data warehouse is a subject-oriented, integrated, time-variant, non-volatile col-
lection of information that is optimized for monitoring and analysis in support of
management's decision-making processes [7]. Hospitals have been slower to adopt
such technology than private enterprise, but it is now an active focus area [2]. Typi-
cally, the user interface for accessing the reporting, monitoring, and analysis functio-
nality of the performance management system is a browser-based "portal". The "por-
tal" view can be customized for each user to define their particular business view of
enterprise data and highlight the individual reports and analyses that are most relevant
to them. A hospital performance management system presents special challenges for
quality assurance. In our case study we have considered the following:
Architectural Complexity of Enterprise Performance Management
Simplistic testing at the level of the user interface can detect quality issues, but it is
not sufficient to determine root causes given the system's complex architecture.
Moreover, each service in the architecture typically has detailed logs [10]. To leve-
rage these, though, a mechanism is needed in order to correlate such a wide variety
of data sources with quality issues experienced by users.
Dynamic Change that affects Quality
Even though it is assembled from off the shelf components, the quality of such a
system (usefulness, performance, reliability, etc.) is affected by the volume and
structure of data, and the specifications of both user interface "portals" and the re-
ports viewed in them. Quality assurance must be managed in production the same
as when testing.
Sensitivity of Data (privacy)
Often the data in the production environment is highly sensitive and access to it is
strictly regulated by privacy law. To accurately assess quality in a test environ-
ment, there must be a mechanism to generate "test" data that is not sensitive but
still exhibits similar characteristics to the "real" sensitive data when running re-
ports.
In addressing these issues in our health performance management case study, we have
prototyped a framework for collecting log files into a data warehouse in order to
leverage the facilities of the performance management system itself to monitor, report
and analyze quality. We have also developed a tool to generate "test" non-sensitive
data that will exhibit the same performance characteristics as "real" sensitive data.
2 Case Study
The hospital in our case study is one of Canada's bigger teaching hospitals with
10,000 employees and over 1,000 physicians. In a typical year, there are over 100,000
emergency visits and 10 million lab tests. The data warehouse has over 100 million
records, and adds tens of thousands of new records a week [4].
For privacy reasons, we are unable to have access to the "real" data in production.
Instead, we generate "test" but similar data in comparable volumes. The data base
schemas are the same. The off the shelf components used are either the same or are
being considered for introduction into the production environment. We use a testing
tool, OpenSTA [9], to automate test scripts and simulate up to 50 concurrent requests
130
over extended periods of time running reports to simulate the load that must be han-
dled in production. Our focus has not been on testing the accuracy of the reports, but
rather on testing the quality of service provided by the system in terms of:
Usefulness - how much of the system is used by which users with what
frequency.
Performance - average response time under different levels of load.
Reliability - monitoring system status and component failures over time under
normal and extreme load.
3 Quality Assurance Framework
Fig. 1. Quality Assurance Framework.
Figure 1 shows the integration of our quality assurance framework with the SOA of
the performance management system. The OpenSTA client simulates user actions in
the Enterprise PM Portal. There are log files that contain detailed information from
each component of the system. The OpenSTA log records the results and execution
time of each user action, the Enterprise PM Portal and each service in the SOA has a
log of actions performed. In addition, there is a database log and a server log for each
physical box that monitors memory usage, file handles etc. The QA Log Loader
processes these files into a QA Data Mart that can be used to report and analyze the
results of any test run. Associated with the QA Data Mart is a QA Portal (Figure 2) to
access the catalogue of QA reports, along with tools to analyze and monitor what is
taking place, making it very easy to notice a quality issue and investigate the cause of
it [8].
The result is that members of the QA team can do a deeper analysis of quality. For
example, in one test run, the OpenSTA log showed slower than normal response
times, but there were other reports to indicate if the database or the report service was
slow, and yet another report that showed a particular server were running out of
131
Fig. 2. Quality Assurance Portal.
memory. In this manner, the performance management system itself becomes a tool
for processing log data and communicating the results of test runs. Another advantage
of this approach is that the quality assurance framework is embedded into the perfor-
mance management system and can remain there in production. This is important
because quality characteristics can change as data and user volumes increase, or as
new reports or portal configurations are added to the system.
There are a number of challenges with this approach though. It is dependent on
the existence and accuracy of the logs provided by the different components of the
system. It can also be very challenging to correlate log entries to individual user ac-
tions at the interface level. At this time, we are only analyzing aggregate results of
overall system quality. The approach also requires skills in data warehouse tools.
Finally, there is a performance and behavioral overhead directly proportional to the
amount of logging that is enabled. Log too much data and the quality of the system is
impacted.
4 Generation of Test Data
To provide an accurate assessment of quality, we had to ensure that "test" data used
was similar to "real" data in terms of both volume and actual values so that reports
run in our test environment would have similar processing characteristics as in pro-
duction. For privacy reasons, "real" data could not be used for testing. As well, data
generated for a data warehouse has a very specialized structure called a star schema
[7], in which data is organized into dimensions (Doctor, Diagnosis, Date, Prescription
Drug, Lab Test) and quantifiable facts (Prescription Amount, Cost, Length of Stay).
In particular, for a performance management system, data and reports are very much
time-oriented.
132
In our approach, we first create by hand a series of very small data marts with fa-
bricated but representative facts using the exact same dimensions as the production
data warehouse. Then, we generate a full volume "test" data warehouse by obfuscat-
ing the identity information in the dimensions, and using a data generation tool to
populate "test" facts from the very small data marts. For example, we built a very
small data mart for antibiotic prescriptions during one week in January 2002 and
cloned the data to fully populate a data warehouse from January 2002 to December
2006 specifying for each month in that time period what the total number of prescrip-
tions should be and the average amount of each prescription. Rather than specific
names for drugs, doctors or diagnoses, "test" names were generated like Drug1, Doc-
tor2, and Diagnosis.
We evaluated some popular test data generation tools [1, 3, 11]. All of them could
generate large volumes of data, one table at a time, using random data or predefined
data sets. Some had adequate support for obfuscating data.
Unfortunately, none had specific support for linking generation of fact table en-
tries to dimensions in a star schema. Nor did any have specific support for time based
generation to fill out data on a month by month basis. Most importantly there was no
support for ensuring that generated data matched aggregate targets on a monthly
basis, or match a specified distribution model. As a result, we have built our own tool
that we have successfully used to generate data for 3 different data marts. It has sup-
port for large volumes of data, obfuscation, star schema and aggregate targets on a
time period basis. The support for aggregate targets on a time period basis is a bit
simplistic, but work is progressing on a general expression engine that can be used to
specify distributions.
5 Results
We evaluated our quality assurance framework in comparison with traditional black
box testing (using a tool such as OpenSTA). There are two aspects to consider in
evaluating a quality assurance framework. One aspect is the capabilities provided in
terms of what aspects of quality assurance can be measured and the quality issues the
framework address. We focus on the scalability, reliability, usage and privacy of the
performance management system, since those qualities have the biggest impact on
whether or not the system would be successful. The other aspect is how much effort
is required to implement and maintain the quality assurance framework, and how
much effort is needed to notice or investigate quality issues, as well as who can do it
and how much skill and knowledge do they need. We also evaluated tools used in
generating "test" data when sensitive "real" data is not available in a test environment.
5.1 Quality Assurance Capabilities
The criteria used for evaluating capabilities of a quality assurance framework for
performance management systems are as follows:
133
Support for assessment of non-functional qualities like scalability, reliability,
usage and privacy of the performance management system.
The ability to analyze detail level information of data collected.
Support for aggregations on quality data with drill up/down in different di-
mensions such as time, services, etc.
Support for alerts to automatically notify when data collected indicated a
quality issue
The ability to perform quality assurance across the entire architecture of the
performance management system on a component by component basis.
The ability to support ongoing quality assurance of the performance man-
agement system in production.
The following table shows the summary of the evaluation of black box testing with a
professional tool and the integrated quality assurance framework in doing quality
assurance for the hospital performance management system in our case study.
Table. 1. Quality Assurance Criteria.
Criteria Black Box Testing Integrated Framework
Ensure
non-functional qualities
Scalability yes yes
Reliability yes yes
Usage no yes
Privacy no yes
Detail Results
yes yes
Aggregate Results with Drill up/down
no yes
Alert system quality exceptions
no yes
QA across individual SOA components
no yes
Quality assurance in production
no yes
Black box testing with OpenSTA can be used to simulate real user behaviors to the
performance management system and report HTTP feedbacks returned by web serv-
ers with graphs and tables showing detail information of data collected. Useful re-
ports are provided to show scalability and reliability of the performance management
system on the whole system level.
But black box testing does not access the underlying architecture of the perfor-
mance management system; hence qualities of components/service inside the system
infrastructure can not be tested and deep analysis on the system qualities can not be
done. It can not ensure usage and privacy of the system due to lacking of related data.
It only provides simple aggregations on data. There are no mechanisms provided to
alert users of system exceptions. Users have to run and view reports manually after a
test run is completed. Finally, simple black box testing is inappropriate for managing
quality assurance of the system in production.
134
In contrast, the integrated quality assurance framework is fully integrated into the
performance management system and across the system architecture. It collects and
consolidated logs that track system and component behavior into the data warehouse
and models quality data in multi-dimensional schema that allow users to drill
up/down through a detailed picture of quality assurance along different dimensions. It
supports the management of non-functional qualities like scalability, reliability and
usage, as well as the ability to monitor privacy of the performance management sys-
tem. As well, the underlying alert mechanisms of the performance management sys-
tem can be used to automatically generate alerts for quality assurance issues. For
example, we can use email function of the system to notice a new report ready for
users, or we can set up quality metrics and their ranges and make the system monitor
those metrics against data and send email for data out of range. Finally, the integrated
quality assurance framework can be used in production to continuously collect and
monitor quality assurance data although this feature may need tuning and configura-
tion as to how much log information is collected.
5.2 Implementation Effort
The criteria for evaluating implementation effort associated with a quality assurance
framework for performance management systems are as follows:
Efforts to implement the quality assurance framework
Efforts to maintain the quality assurance framework
Efforts to notice a quality issue
Efforts to investigate and determine cause of an issue
Additional efforts to perform quality assurance in production
Efforts we used in our experiment to perform quality assurance for the hospital
performance management system are shown in Table 2.
Table 2. Effort Comparison.
Criteria Black Box Framework
implement
1 person day 1 person month
maintain
1 person day 1 person week
notice an issue
scalability
View reports Alert to view reports
reliability
View reports Alert to view reports
usage
Can not be done Alert to view reports
privacy
Can not be done Analyze reports
determine cause scalability
Inspect distributed logs Analyze reports.
reliability
Inspect distributed logs Analyze reports.
usage
Can not be done Analyze reports
privacy
Can not be done Analyze reports
Production Can not be done Built In
135
In the black box testing framework, we do not need to be familiar with the infra-
structure of the performance management system as we only target the web pages of
the Enterprise PM Portal. There are efforts needed to analyze user scenarios and
record user actions as test scripts. The test scripts were saved in OpenSTA project
files and copied between client computers. Test scripts can be replayed by client
computers to create required traffic on the system, and then reports are automatically
created after test runs for analyzing collected data. This part was finished in one day.
The integrated QA framework, on the other hand required lots of effort to collect
log data into the data warehouse. First, we needed to enable log functions for system
components. Second, we needed to analyze logs and implement a log loader for them.
The ETL process involves running the log loader and executing database SQL com-
mands or functions, and double checking the data collected to ensure the quality of
the data collected. Third, we used a metadata modeling tool to model collected data
dimensionally in the data warehouse and publish a metadata model for reporting.
Finally, the quality assurance reports and portal needed to be designed and created
manually. It took one month to set it up.
In terms of maintenance for the black box testing framework, changes on web pag-
es will cause OpenSTA to redo the work of recording and replying test scripts, but it
is still quite simple and can be done in one day.
For the integrated quality assurance framework, significant maintenance efforts are
needed when there are changes in the system infrastructure, for example:
Changes in applications or services for the performance management sys-
tem will bring different logs into the quality assurance framework. The
log loader has to be compatible with those logs and the metadata model of
the quality assurance data mart may need to be changed too.
Changes on the granularity of logs of a component will cause configura-
tions on the log loader and related changes between metadata objects in
the quality assurance data mart.
Changes on the goal of quality assurance may ask for new reports or mod-
ification on existing reports, or the quality assurance portal.
In general, all of this work can be done in one week. It can be faster if the inte-
grated framework is designed to be compatible with different service and component
configurations.
Implementation and maintenance effort is more significant with the integrated
quality assurance framework, but once in place the effort associated with assessing
quality assurance is greatly reduced. With the black box testing framework, the effort
needed to notice a quality issue is usually high. First, there is no alert provided to
users for system exceptions, users need to discover them manually. Second, although
a simple reporting capability is usually bundled into black box testing tools, there is
typically no relationship set up between reports and there is no portal to help users
easily go through reports. To find quality issues, users need go through all reports and
figure out the data relation between reports; this is difficult even for quality assurance
experts. As well due to the limitation on the type of data collected, there is no support
for assessing quality issues such as the usage and privacy of the system.
Reports in the integrated quality assurance framework are created based on the
quality assurance data mart model, a star schema, and reports content can be custo-
136
mized to allow users to easily find out data and graphs that they are interested. Users
can drill up and down in dimensions to see the system qualities in different levels and
drill though different reports to analyze related information. The quality assurance
portal can also be customized by users to organize reports, which help users to focus
on their goals. The integrated quantity assurance framework can notice issues on
scalability, reliability, usage and privacy of the system.
The black box testing framework cannot perform quality assurance across the per-
formance management system architecture; hence, it cannot report what happened
inside the box. To investigate and determine causes of an issue, users need manually
inspect system components’ logs distributed in the system. Those logs often have
complex structure, complex information and very long content, and users have to
look through the content line by line. To do this, time and efforts tend to be huge.
Without related data collected, it cannot investigate and determine cause of a usage or
privacy issue of the system
The integrated quality assurance framework can perform quality assurance across
the performance management system architecture. The framework can collect data
from logs automatically by the ETL process, so data in the quality assurance data
mart is well structured, cleaned and meaningful. The quality assurance portal allows
users to view quality data in a single place and reports let users find the relevant data
very quickly. The integrated quality assurance framework can report issues on scala-
bility, reliability, and usage of the system. The integrated quality assurance frame-
work does not provide an automatic way to flag privacy violations, but it does pro-
vide usage analysis reports from audit logs that can be used to discover and investi-
gate potential privacy violations.
In general, while the integrated quality assurance framework requires greater effort
and cost to implement, it does provide significant benefits and requires less cost and
effort to use both in a test environment and in production.
6 Conclusions
The initial results of this case study are promising. We can generate "test" data in
volume to assess the quality of a hospital performance management system in a test
environment. Through careful analysis of system architecture and log files coupled
with data engineering of a quality assurance data mart, we have been able to create an
integrated quality assurance framework that leverages the performance management
system to provide deep analysis of quality at an aggregate level which is available
both in the test and production environment. In future work we hope to support deep
analysis at a detail level, and provide even more sophisticated data generation fea-
tures.
References
1. Advanced Data Generator 1.7.0, Upscene Productions, http://www.upscene.com/, last
retrieved: 2008/04
137
2. P. Chountas, V. Kodogiannis, Development of a Clinical Data Warehouse. Medical Infor-
mation Systems: The Digital Hospital (IDEAS-DH'04), pp 18-14, September 2004. ISBN:
0-7695-2289-0
3. DTM Data Generator 1.15.04, DTM Soft, http://www.sqledit.com, last retrieved: 2008/04
4. A.J. Forster, The Ottawa Hospital Data Warehouse: Obstacles and Opportunities. IT in
Healthcare Series, June 15, 2005
5. Y.F. Jarrar, A. Al-Mudimigh, M. Zairi, "ERP implementation critical success factors -the
role and impact of business process management", IEEE International Conference on Man-
agement of Innovation and Technology, 2000
6. J.J. Jeng, "Service-Oriented Business Performance Management for Real-Time Enterprise",
E-Commerce Technology, 2006.
7. R. Kimball and M. Ross, “The Data Warehouse Toolkit: The Complete Guide to Dimen-
sional Modeling”, Second Edition, Wiley, 2002.
8. L. Peyton, A. Rajwani, "A Generative Framework for Managed Services", International
Conference on Generative Programming and Component Engineering, October, 2004.
9. OpenSTA (Open System Testing Architecture), http://www.opensta.org, last retrieved:
2008/04
10. C. Rankin, The Software Testing Automation framework, IBM Systems Journal, Software
Testing and Verification, Vol. 41, No.1, 2002
11. TurboData 4.0.6, Canam Software, http://canamsoftware.com,last retrieved: 2008/04
12. W3C Working Group, Web Services Architecture, Note 11 February 2004, http://
www.w3.org/TR/ws-arch, last retrieved: 2008/04
138