Project

Database Design for the Next Generation of Networks

Principal Investigator

Tim Kraska

Brown University

Oracle Principal Investigator

Garret Swart, Architect

Summary

The current trend towards high-performance RDMA capable networks, such as
Infiniband FDR/EDR, will fundamentally change the design of modern database
systems. Current distributed database systems are build on the assumptions,
that network communication is slow and that network bandwidth is severely
limited. With the next generation of networks this assumption no longer
holds; the whole cluster becomes more balanced and it is no longer sufficient
to optimize all algorithms around this single bottleneck.

The goal of the project is to systematically design, implement and evaluate
transaction execution strategies for the next generation of network technology,
such as Infiniband EDR or Ethernet 100G. Most importantly, we will develop a
new RDMA-enabled snapshot isolation protocol, which avoids the typical large
overhead for distributed transactions, and achieves much higher throughput
rates than more traditional alternatives.