Back to search

Boston University

From page importance to memory characterization: profile-driven heterogeneous memory management in high-performance embedded systems

Abstract

dc:description.abstract

Modern embedded multicore SoCs increasingly combine (i) tightly shared cache and memory infrastructure and (ii) heterogeneous memory resources spanning diverse technologies and access paths. In real-time and mixed-criticality settings, these trends undermine predictability: execution time becomes a function not only of an application’s own memory behavior, but also of interference caused by co-running tasks and devices, while the growing diversity of memory resources expands the design space for page placement and resource isolation. This dissertation argues that predictable and efficient memory management on such platforms requires moving beyond blind, capacity- and locality-centric allocation policies toward profile-driven orchestration. To provide the foundational visibility needed for this approach, the thesis contributes two complementary toolchains. First, BBProf provides demand-side visibility by extracting per-page temporal importance. It ranks an application’s pages by their end-to end timing impact, allowing system software to separate frequently accessed but low-consequence pages from pages whose slowdown or eviction cause disproportionate performance collapse. Second, MemScope provides supply-side visibility via a kernel-level characterization framework for heterogeneous memory subsystems. MemScope coordinates stressor and observer cores in kernel-space and runs experiments under controlled interference. This enables repeatable extraction of bandwidth, latency, and contention sensitivity across memory technologies and access paths, including cross-path interference driven by the shared on-chip interconnect. Building on these results, the dissertation outlines a roadmap for MatchMaker, an OS-level framework that closes the loop by combining BBProf’s per-page importance with MemScope’s per-memory module characterization to guide allocation and isolation decisions toward improved predictability and performance under contention.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy
Level thesis:degree_level
doctoral
Discipline thesis:degree_discipline
Electrical & Computer Engineering
Grantor
Boston University
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Ghaemi, Golsana
Advisor dc:contributor.advisor
  • Mancuso, Renato
ORCID dc:identifier.orcid

Subjects

dc:subject × 1

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/2144/53034

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Ghaemi, Golsana. From page importance to memory characterization: profile-driven heterogeneous memory management in high-performance embedded systems. doctoral thesis, Boston University, 2026. https://hdl.handle.net/2144/53034