Clustering TDB, Canberra, 9 July 2007

Recommended:

DCS SEMINAR SERIES

Department of Computer Science
The Australian National University
Canberra

Clustering TDB - a Little Database Meets Big Iron
Andrew Tridgell (IBM OzLabs)

DATE: 2007-07-09
TIME: 16:00:00 - 17:00:00
LOCATION: CSIT Seminar Room, N101

ABSTRACT:
The little tdb database has been at the heart of Samba for many years. More recently it has become a focus in the efforts to create a clustered Samba solution. Attempts to cluster Samba in the past have been patchy at best, oftentimes resulting in a solution which relies on "relaxed" data coherence rules, which is a polite way of saying that they can lose user data. In the last few months we have finally worked out a architecture for a clustered tdb which will allow us to achieve three goals simultaneously:

1) scaling to large numbers of nodes in a cluster (potentially into the hundreds)
2) robustness to node crashes and new nodes entering the cluster
3) correct data coherence rules, so user data is not at risk
4) High Availability, including IP takeover, service management and related tasks

This talk will present the solution we have come up with, and will discuss our experiments and the results of the first production installation.

BIO:
Andrew Tridgell is a free software developer working for IBM OzLabs from his home in Canberra.

Tridge is widely known as the inventor of Samba and as a key contributor to Linux development. He has had a long association with ANU both as a student (having earned both his BSc and his PhD here) and a researcher in operating systems for highly parallel machines.

ps: I assume this is a rerun of Tridge's talk at the January Linux Conference.