Back to search

Cornell University

Towards High Performance Abstractions for Strong Geo-Replicated Systems

Abstract

dc:description.abstract

Large-scale Internet applications have become ubiquitous in everyday life because theyseamlessly and persistently connect people to each other and to service providers. These applications are built on geo-replicated services that reliably process and store user data. Application developers prefer services that provide strong guarantees because they simplify the process of programming correct applications. However, existing approaches to implementing geo-replicated services that meet the high throughput and low latency requirements of large-scale Internet applications sacrifice strong guarantees or generality in the application programming interface. This dissertation argues that geo-replicated services that provide strong guaranteescan meet the high throughput and low latency requirements of large-scale Internet applications without sacrificing generality in the API. To do so, we propose that services more effectively leverage semantic information already present in existing, general APIs. We demonstrate this approach in the design of two systems, Morty and Gryff. Mortyis a replicated transactional storage system that provides serializable transactions with high throughput under contention in geo-replicated settings. Morty achieves higher throughput than existing systems by leveraging a continuation passing style API—which is already commonly used in networked services—to implement transaction re-execution. Gryff is a replicated coordination service that provides linearizability with low read tail latency in geo-replicated settings. Gryff achieves lower read tail latency than existing systems by processing simple reads and writes with a shared register protocol instead of consensus. Because coordination services already differentiate between reads and writes and stronger synchronization operations like read-modify-writes, developers can leverage Gryff’s more efficient design to provide lower latency to end-users without significantly rewriting their applications.

Degree

thesis:*
Name thesis:degree_name
Ph. D., Computer Science
Level thesis:degree_level
Doctor of Philosophy
Discipline thesis:degree_discipline
Computer Science
Grantor
Cornell University
Year dc:date.issued
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Burke, Matthew
Chair dc:contributor.chair
  • Alvisi, Lorenzo
Committee members dc:contributor.committeemember
  • Agarwal, Rachit
  • Weatherspoon, Hakim

Subjects

dc:subject × 6

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
16575510
Dc Identifier Other
ProQuest Submission ID: 14152
ProQuest Publication ID: 30991750

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Burke, Matthew. Towards High Performance Abstractions for Strong Geo-Replicated Systems. Doctor of Philosophy thesis, Cornell University, 2024. https://hdl.handle.net/1813/115894