Parallel Architecture Optimization for Threaded Applications

Abstract

As computers today move away from the local domain and into the cloud, it is imperative to have data center architectures that can reliably and efficiently deal with the increasing demand. A new architecture called the Sharing Architecture aims to decentralize the long-living virtual core to give clients the ability to fully customize the resources they individually need for their applications. In order to further push the Sharing Architecture to functional usage, it is important that the new design supports multi-threading, a key concept and requirement for today’s computing needs. This paper looks into the concept of threading as it applies to general architectures, and attempts to model a solution that takes into account data dependencies and timing slippage between a functional and performance simulator. This paper discusses several approaches for dealing with multi-threaded execution on the Sharing Architecture, and describes in detail a new method for extending the Sharing Architecture to support multi-threaded programs.