Transactional Partitioning: A New Abstraction for Main-Memory Databases

TL;DR

This paper introduces a new high-level programming abstraction called transactional partitioning for main-memory OLTP databases, aiming to improve performance, scalability, and ease of development by enabling developers to reason about and influence program performance.

Contribution

It proposes a novel transactional partitioning model that extends ACID transactions, allowing application developers to optimize performance without database designer involvement.

Findings

Supports scalable, high-performance OLTP applications on commodity hardware

Maintains full ACID semantics within the new programming model

Enables developers to influence performance directly

Abstract

The growth in variety and volume of OLTP (Online Transaction Processing) applications poses a challenge to OLTP systems to meet performance and cost demands in the existing hardware landscape. These applications are highly interactive (latency sensitive) and require update consistency. They target commodity hardware for deployment and demand scalability in throughput with increasing clients and data. Currently, OLTP systems used by these applications provide trade-offs in performance and ease of development over a variety of applications. In order to bridge the gap between performance and ease of development, we propose an intuitive, high-level programming model which allows OLTP applications to be modeled as a cluster of application logic units. By extending transactions guaranteeing full ACID semantics to provide the proposed model, we maintain ease of application development. The model allows the application developer to reason about program performance, and to influence it without the involvement of OLTP system designers (database designers) and/or DBAs. As a result, the database designer is free to focus on efficient running of programs to ensure optimal cluster resource utilization.