PhD thesis defense to be held on July 10, 2025, at 12:00 (NTUA Administration Building)
Thesis title: ENABLING BEHAVIOR-BASED ENERGY CONSUMPTION AND MEMORY FOOTPRINT OPTIMIZATIONS IN NATIVE CONTEXTS
Abstract:Computers control a large part of modern life despite being a (very) small part of history. And while exotic new applications such as ChatGPT attract most of the interest from both consumers and practitioners, the first principles underlying our world’s digital infrastructure remain timeless. This thesis studies the following such principles through the lens of energy consumption and memory footprint: (i) a central aspect of program behavior is its dynamic requests for memory, (ii) approximately optimal solutions to memory allocation can be computed offline and (iii) software is the result of iterative decision-making over source code transformations. Along the way, we make a series of original contributions. We show the complex impact that specific dynamic memory allocation implementations have on the extremely popular Python programming language; we describe a principled methodology for capturing program-allocator interaction and quantifying memory fragmentation; we contribute a static memory planning implementation outperforming the SOTA in a wide range of heavyweight, challenging benchmarks; and we demonstrate a flexible, agnostic framework for improving software.
Supervisor: Professor Dimitrios Soudris
PhD Student: Christos Lamprakos