Information Technology Reference
In-Depth Information
always straightforward. This is because information about correlator attributes
may not be known to a monitoring service which overseas and captures service
interactions in a log. Although each service internally knows how it correlates
its messages with those of its immediate interacting partners, this information
may not be well documented, or may be buried in the code of the service which
is sometimes outsourced or the documentation may become obsolete. Therefore,
it is often necessary to perform an automated message correlation in order to
identify process instances.
We consider two messages in a service interaction log as correlated if a
Cor-
relation Condition
(CC) is verified. A typical CC may be the equality of two
messages' attributes [8]. When messages generated by the same process instance
are correlated, the sequence of their CCs can be considered as the ”fingerprint”
of such process instance. Therefore, discovering sequences of CCs allows to step
ahead towards the identification of process instances.
Motivated by the goal of providing a light-weight approach that can help an
analyst to quickly identify relevant sequences of CCs, we propose in this paper an
approach using message indexation and aggregations to generate an Aggregated
Correlation Graph (ACG) that exhibits all the sequences of CCs identified in a
service interaction log. In an ACG graph, each node corresponds to an aggrega-
tion of messages, and each edge represents a CC between all pairs of messages
in the two nodes. Therefore, the ACG graph represents all correlations of mes-
sages in an aggregated representation allowing to quickly identify, using weighted
nodes and edges, the most frequent sequences of CCs revealing frequent process
executions. The approach has been implemented and we offer an interactive fil-
tering/browsing of the ACG graph helpful for analysts to better understand the
way messages are correlated and the potential process models those correlations
may reveal. In particular, we make the following contributions:
(1) We propose an approach for service interaction message aggregation based
on their CCs. The resulting ACG graph is size-ecient and exhibits all the
sequences of CCs identified in the log.
(2) We provide a method, based on graph filtering techniques and user-defined
criteria, to eciently identify relevant sequences of CCs and visually browse
them using an interactive ACG graph visualization tool.
(3) We have implemented the approach in a tool available online, and performed
experiments on a number of service interaction logs. The experiments show that
the generated ACG has a stable size regardless of the size of the log being
processed. Furthermore, the interactive ACG visualizer allows analysts to quickly
find relevant sequences of CCs and identify process instances.
The rest of the paper is organized as follows: In section 2, assumptions and
notations used in this paper are presented. We present the service interaction
message log format, and define the notations used in this paper. In section 3, we
propose our approach and discuss its strengths and limitations. In section 4, we
describe the implementation and discuss its results and applications. In section
5, we discuss related work. Finally, in section 6, we summarize the contribution
of this paper and present future work.
