Processing-in-Memory (PIM) for Cloud Computing: Optimizing Performance and Energy Efficiency

Somvanshi, Savitanshu (2025) Processing-in-Memory (PIM) for Cloud Computing: Optimizing Performance and Energy Efficiency. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (949kB) | Preview
Preview	PDF (Configuration Manual) Download (715kB) | Preview

Abstract

Modern cloud systems often struggle to efficiently handle memory-bound workloads, particularly as data-intensive applications become more common. Conventional schedulers, which rely on static rules for job classification and assignment, are unable to adapt to changing workload patterns in real time. This limitation contributes to increased latency, misclassified job placements, and higher energy consumption especially in heterogeneous environments where traditional Central Processing Unit (CPUs) operate alongside Processing-in-Memory (PIM) enabled nodes. To address this, a lightweight and feedback-aware scheduling strategy is proposed. The scheduler operates within a CloudSim-based simulation and avoids fixed heuristics by profiling each job using a brief 10% execution sample. Runtime is estimated and used in combination with a dynamically adjusted RAM-to-Length ratio threshold for classification. A feedback loop monitors prediction error and continuously refines the threshold to improve accuracy over time. The system is evaluated using 100 synthetic cloudlets processed under identical simulation conditions. Key performance indicators such as execution latency, prediction accuracy, and energy consumption are recorded. In experiments, the adaptive scheduler reduced average energy consumption by 15% and make-span by 18% compared to a static baseline, while maintaining more stable latency. Results indicate that the adaptive scheduler reduces both average latency and energy usage while maintaining stable job classification behavior. These findings suggest that lightweight, feedback-driven scheduling strategies can improve resource efficiency in CPU–PIM hybrid environments without introducing significant complexity.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Sahni, Vikas UNSPECIFIED
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Cloud computing T Technology > T Technology (General) > Information Technology
Divisions:	School of Computing > Master of Science in Cloud Computing
Depositing User:	Ciara O'Brien
Date Deposited:	31 Mar 2026 10:27
Last Modified:	31 Mar 2026 10:27
URI:	https://norma.ncirl.ie/id/eprint/9271

Actions (login required)

View Item