λ group: Using Optics to Take Group Data Delivery in the Datacenter to the Next Degree

Description
Abstract

The increasing number of datacenter applications with heavy one-to-many communications has raised the need for an efficient group data delivery solution. This paper presents an unconventional clean-slate architecture called λ group that uses optical networking technologies to enable ultra-fast, energy-efficient, low cost, and highly reliable group data delivery in the datacenter. By replicating data purely optically, λ group enables reliable one-to-many communication independent of modulated data rate, while simultaneously being energy-efficient. λ group makes use of software defined network switches to facilitate direct interactions between applications and the network, eliminating the need for conventional multicast routing and group management protocols without sacrificing link-stress or latency performance. From an end-to-end perspective, λ group provides predictable latency and near-zero packet loss between application end points, allowing even simple end-to-end flow control and packet loss recovery mechanisms to work well. We implement a λ group prototype to measure its physical layer characteristics and end-to-end performance benefits to applications. Extensive simulations using synthetic traffic show λ group provides an order of magnitude performance improvement relative to a number of existing alternative group data delivery approaches.

Description
Advisor
Degree
Type
Technical report
Keywords
Citation

Bergman, Keren, Ng, T. S. Eugene, Sripanidkulchai, Kunwadee, et al.. "? group: Using Optics to Take Group Data Delivery in the Datacenter to the Next Degree." (2014) https://hdl.handle.net/1911/96412.

Has part(s)
Forms part of
Published Version
Rights
You are granted permission for the noncommercial reproduction, distribution, display, and performance of this technical report in any format, but this permission is only for a period of forty-five (45) days from the most recent time that you verified that this technical report is still available from the Computer Science Department of Rice University under terms that include this permission. All other rights are reserved by the author(s).
Link to license
Citable link to this page