Research programs framed around measurable system gains and operational insight.

Each stream is presented through the engineering objective, the test environment, the validation approach, and the quantified results achieved across HPC, distributed, and energy-aware systems.

CPU Power

76 W

GPU Power

122 W

Forecast Load

68%

Workload distribution

CPU58%
GPU42%

Power forecast

Predictive Energy, Performance Optimisation

Across modern computing infrastructure, from cloud platforms to high performance systems, there is a persistent and expensive blind spot. Systems are monitored, but not truly understood.

My work began with a simple but difficult question: can system behaviour be predicted reliably using only measurable signals? To answer this, I built a measurement driven approach that connects CPU activity, frequency scaling, and power behaviour with real workload outcomes. In practice, this delivered execution gains of up to 34%, and moved optimisation from guesswork to repeatable, data driven decision making.

34%

Execution gain achieved

UCD / Linux

Primary test environment

HPCLinuxEnergy OptimisationPredictionLIKWIDIntel PCM
Google Scholar
Distributed control planeReplica orchestration

Ingress

Request Router

Control

Coordinator

Consistency

Quorum Logic

Replica 1Replica 2Replica 3Client APICoordinator

Throughput

+35%

Replica sync

98.8%

Stale reads

-25%

Distributed System Control & Optimisation

Distributed systems are the backbone of modern applications, yet they are often managed rather than truly controlled. When behaviour becomes unpredictable, dashboards and partial metrics are not enough.

Working with Apache Cassandra, I designed a middleware layer that made the environment measurable and controllable from within. By running controlled experiments across nodes, workloads, and configurations, it became possible to explain system behaviour directly, increase throughput by 35%, and reduce stale reads by 25%, turning a black box into something predictable and optimisable.

35%

Throughput improvement

25%

Stale reads reduced

Apache CassandraDistributed SystemsConsistencyMiddlewarePrometheusGrafana
Related publication