Building a network coded storage testbed for data center energy reduction

Abstract

The energy consumption of data centers (DCs) worldwide is approaching that of countries such as Argentina. This paper builds upon the analytical work presented in [1], which showed that reductions in blocking probability, using algorithms such as network coded storage (NCS), can significantly reduce DC energy consumption. We demonstrate that NCS can be deployed in a DC management software with a commercial grade feature set, without interference. We develop a simple method to estimate blocking probability DC networks, and present preliminary blocking results for a single-server single-drive node. The next step of this project is to use the insights gained herein to develop more accurate queueing models for delay and blocking characteristics of DC servers and drives.

Scheduling advantages of network coded storage in point-to-multipoint networks

“…2 We will discuss the relation between external and internal blocking in Section II-D. This is a similar model to existing drive blocking models [1] and existing practical server experimentation test [15]. When a user request arrives and is not externally blocked, one slot of the server buffer is allocated to manage and service this user request.…”