Database Reference
In-Depth Information
that non-termination may occur. We therefore advise the user that either C7 or C12 must
eventually become false in the specifi cation. Otherwise, there is a fundamental fl aw in the
workfl ow; that is, the cycle may never terminate.
Algorithm 1 provides a means to detect and correct non-terminating cycles
. However,
it provides much more. It forces us to analyze the statechart in a systematic manner. Our fi rst
analytical action after we fi nished design of the MWI workfl ow statechart was to identify
cycles in the workfl ow. Once all of the cycles were identifi ed, we then began to look closely
at each cycle for possible non-termination. Without Algorithm 1, we would never detect or
even suspect non-termination problems. Thus, Algorithm 1 acts as a high-level analytical
tool to systematically identify and correct non-termination problems within a given stat-
echart. System designers can identify, discuss, and correct potential cycle non-termination
problems during design rather than attempt to correct problems in production. Of course,
system designers can also correct non-termination problems for existing systems in the
manner discussed in the MWI case.
Another potential workfl ow problem is confl uence.
Algorithm 2 provides a means to
identify a set of transitions that may lead to non-confl uence
. In Figure 10, we identifi ed an
and-state in the sub-state Labeling contained in the high-level state Scan/Bundle card. Ac-
cording to Harvey Black, sometimes package labels are put on bundles or bundle labels are
put on packages. Thus, the and-state Labeling may have a write-write racing problem.
Notice that we spoke with Mr. Black once we noticed the and-state. After Mr. Black was
made aware of the and-state, he was able to better understand where the workfl ow problems
were occurring. Of course, we didn't explain the details of statecharts to Mr. Black. We
instead explained to him the situation in business language. As a result of our intervention,
MWI is attempting to rethink the labeling process.
Hence, Algorithm 2 provided a systematic basis for redesign. By using the principles
developed in Algorithm 2, we were able to fl ag the and-state structure as a source of potential
problems and inform the user about it.
Another potential workfl ow problem is observable determinism. As Theorem 5 in-
dicates, if a statechart is not confl uent and its outputs are visible, then it is not observably
deterministic. Since the and-state Labeling has a write-write racing problem and its outputs
are visible to the environment, the outputs of the and-state Labeling are not observably
deterministic. After explaining to Mr. Black the concept of observable determinism, he was
able to identify a potential workfl ow problem. During label printing, there is a real danger
that package labels and bundle labels can be switched. Although it is easy to distinguish by
eye the difference between a bundle (a set of packages) and a package, it is very possible
that an operator will accidentally place a bundle label on a package and vice-versa. Keep
in mind that the labels are both plain white and the bar codes are not easy to see with the
naked eye. As a result of this analysis, Mr. Black has suggested to management that bundle
and package labels be made different colors. Although color printing is more expensive, the
reduction in errors should more than justify the investment.
Algorithms 1 and 2 offer a systematic means to identify problems in complex busi-
ness workfl ows. By using the principles developed in this study, one can scan any statechart
quickly and effi ciently to fl ag potential workfl ow problems. Process improvement and
redesign has tended to focus on correcting, streamlining, and/or completely rethinking ex-
isting business workfl ows to reap vast improvements in performance and signifi cant costs
savings. However, this study pioneers the use of statechart analysis to identify workfl ow
problems during redesign.
