Network Coding for Strong Consistency Semantics in Distributed Shared Memory Networks
Abstract
The abstraction of shared memory is a powerful tool that can greatly simplify design and implementation of software systems for distributed settings. The key property of a shared memory abstraction is the consistency guarantees that it provides under concurrent access. The most intuitive consistency model is atomicity since its semantics closely reassemble these of a sequential register. Emulating shared atomic memory proves to be challenging especially if this is to be done in a distributed message passing setting with unreliable components, as is often the case in networked systems. To this end standard techniques and solutions have been developed, however these have high communication and storage costs. In this project, we explore several algorithms for efficient implementations of shared memory services that use erasure-codes. We also derive provable strong consistency and liveness properties of these algorithms in the failure-prone setting. Novelty of this work is the fusion of standard replication techniques with network coding techniques. Shared memory services for distributed systems emulate a memory abstraction providing users with a consistent view of its content under concurrent access, with a best effort to ensure availability and survivability.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 14, 2018
- Accession Number
- AD1085863
Entities
People
- Nancy Lynch
Organizations
- Massachusetts Institute of Technology