Estimation of timings for RCASE algorithm to optimise distribution
calculation strategy
Background
Warwick Analytics provides ground breaking software that identifies root causes of faults and
inefficiencies in manufacturing industries. The main algorithm (RCASE) was developed from over 10
years academic research and spun out of Warwick University.
RCASE requires a failure marker (usually from a warranty claim or test case) and some data from the
life cycle of the product – this could include tolerance data, production data, testing data and user
data. It does not require hypothesis and provides results even with dirty/incomplete data sets.
In 2013, Warwick Analytics won Demo God at Demo Fall and SAP’s worldwide most innovative start
up as well as raising initial investment from Jensons Solutions.
Project
RCASE calculations can vary from seconds to days! The algorithm is implemented in 7 distinct
segments each of which can be parallelised. RCASE can be deployed on Windows Azure platform and
spin up as many machines as required to perform the calculations.
The bottlenecks within the algorithm have not been analysed. However, it is confirmed that
bottlenecks occur throughout the algorithm depend on the makeup of the data. There is a
performance hit when distributing an algorithm segment (spinning up machines, splitting the data,
setting up queues, merging the results).
The project is to find a concurrency strategy for the RCASE algorithm based solely on the
characteristics of the input data. By analysing various input data sets and an understanding of the
algorithm, it should be possible to model the relative timings of each segment of the RCASE
algorithm. By using this model, the optimal concurrency strategy can be inferred.
Deliverables
WA will provide access to the RCASE implementations, example data sets and the algorithm
developers. The student is not expected to program but take metrics from the algorithms and come
up with a method for optimisation.
The deliverables are expected to be
1. Data analysis strategy and categorisation
2. Relative timings for each segment of RCASE based on Deliverable 1
3. Optimal distribution strategy of RCASE based on Deliverable 2