DATE: Thursday , November 7, 2002
TIME: Noon - 1 pm
PLACE: Wean Hall 8220
SPEAKER:
Richard Golding
Panasas, Inc.
TITLE:
The Palladio Project:
A Survivable, Scalable Distributed Storage System
ABSTRACT:
Large-scale distributed storage services for cluster computing applications
and for corporate IT systems must both give high performance and high
reliability -- and increasingly, reliability includes the ability to survive
site disasters. Palladio is a research storage system architecture that
addresses these needs through a suite of protocols for run time data access
and system management.
This talk first gives an overview of Palladio, and then focuses on the data access and metadata/structure management protocols. The data access protocols use an optimistic concurrency control technique to provide serialized, atomic reads and writes; simulation results show this approach has higher performance than locking-based mechanisms. The data access protocols interact with the layout control protocols, which handle failure detection, metadata management, and coordination of changes of the layout of data onto storage devices.
BIO:
Dr. Richard Golding is currently a software architect with Panasas, working
on high-performance file server systems. Previously, he was with the Storage
Systems Program at Hewlett-Packard Labs, where he
worked on the AutoRAID project, storage system self-management, and the
Palladio project, as well as working as a technology transfer liaison
to the HP storage system product division. His PhD is from UC
Santa Cruz, having worked on group communication and weak-consistency
wide area systems. Before that he spent a decade doing IT and graphics
applications, including coauthoring the Sun olwm. His research interests
are in reliable and well-performing distributed systems, with a focus
on how decentralised, cooperating nodes can collectively produce a correct
self-managing system.
SDI / LCS Seminar Questions?
Karen Lindenfelser, 86716, or visit www.pdl.cmu.edu/SDI/