Replication for the Global Data Plane

Distributed Replication for a Specialized Database

Scott Numamoto - Tony Yang - Steven Wu

A very rough paper is written about the work we have done. However, the entire project is still in progress.

Abstract

There has been an explosion in the number of smart devices connected via the Internet to augment people's everyday life. The Global Data Plane (GDP) is a new infrastructure design which tries to address the challenges presented by such a vast and broad array of devices by introducing a higher layer of abstraction. We present an anti-entropy mechanism which ensures that data is replicated on multiple GDP log servers and user APIs. We modeled many of our decision to support a highly write available system. Evaluations show that our replication system can recover from large loss of log servers and has significantly better usage of bandwidth than a naive system.

Replication for the Global Data Plane is the result of a course paper for CS 262: Advanced Topics in Computer Systems.

Packages Summaries:

• logserver provides access to the functionality of a GDP log server. We have simulated a log server with a SQLite3 database.
• loggraph provides an abstracted view of the records in the log server as a graph with the ability to read and write records.
• policy dictates what replicas communicate with each other to determine what records to serve.
• peers abstracts how replicas commuicate data with each other
• daemon when to send heartbeats with peers and who to send them to