Abstract:
Process Mining consists of extracting valuable information from event logs
produced by Process Aware Information Systems (PAIS) which support
business processes and generate event logs as a result of execution of the
supported business processes. Alpha Miner is a popular algorithm in Pro-
cess Mining which consists of discovering a process model from the event
logs. Discovering process models from event logs is a computationally in-
tensive and time consuming task in context to processing large volumes
of event log data. In this work, we present a parallel version of the Al-
pha Miner algorithm and apply different types of parallelisms (implicit,
explicit, GPU) provided by MATLAB (Matrix Laboratory). To improve
the program’s performance, we identify its bottleneck and apply implicit
parallelism on it through multithreading done by using arrayfun construct
which perform element wise operation. For explicit parallelism, we use the
parfor construct. We identify independent and computationally intensive
for loops in the Alpha Miner algorithm on which parfor can be applied.
We measure the extent of speedup achieved by implicit and explicit paral-
lelism with respect to serial implementation of Alpha Miner algorithm on
Central Processing Unit (CPU). We compare the performance obtained by
implicit parallelism and explicit parallelism on CPU. Further, we use Graph-
ics Processor Unit (GPU) to run computationally intensive parts of Alpha
Miner algorithm in parallel. On GPU, we do parallelism using arrayfun
construct. We measure the speedup achieved using GPU with respect to
the same program run over multi-core CPU. Alpha Miner algorithm is ac-
celerated the most by GPU with speedup reaching till 39.3×. To test the
efficiency and scalability of different types of parallelisms, we perform tests
on real world as well as synthetic datasets of varying sizes.
