The University of Arizona, Department of Computer Science

Swarm Storage Server

Overview

Swarm is a network storage system that uses log-based striping to achieve high performance and scalability. Clients collect application writes into an append only log and the log is striped across multiple storage servers to obtain performance that is a function of the number of storage servers. The Swarm Storage Server is the component of Swarm that serves file data. The Swarm Server uses the Logical Disk (LD) to manage the disk subsystem.

Research Goals

• Storage Servers. The design goal of the Swarm Server is to effectively satisfy Swarm requirements. These include the need for high write performance, for small as well as large writes, support for security and atomic interface operations.
• Paths. We are studying the use of various path-based optimizations to achieve high performance. We are also looking at using end-to-end paths to explore issues with flow control and providing bandwidth guarantees for clients.

Router Graph

There is one path per TCP connection between the Swarm Stub router and the network. These paths map one-to-one to paths between Swarm Stub and LD. Currently all client calls are synchronous and the state of an outstanding client call is stored in the corresponding path. The future goal is to adapt the current path architecture to support asynchronous calls.

Status

• We have a working prototype that implements all functionality Swarm clients require.
• Our prototype uses approximately 70% of the raw disk bandwidth for writes.

Papers

Contributors

• Rajesh Sundaram, Graduate Student