Next generation data centers: trends and implications

Abstract

In this talk we will discuss next generation data centers and the important impact they will have upon enterprise applications. Specifically, we will discuss the technical and economical trends motivating the move towards large scale distributed data centers consisting of tens of thousands of servers and hundreds of petabytes of storage. We will explain the roles and advantages of virtual machines and other virtualization technologies in these environments and also explore how they exacerbate the complexity of management and achieving predictable application behavior.To better illustrate issues emerging in such environments we will describe early experiments we conducted with a 1000-processor utility rendering service created for DreamWorks Animation that was used to render the films Shrek II and Madagascar. We will discuss the lessons learned from this experience.Next, we consider the trend towards service oriented architectures for enterprise application platforms. Service orientation provides for more flexible and agile information technology systems but further increases the complexity of management and behavior. We will explore the implications of composing services dynamically using an SOA approach.These trends for enterprise application platforms and the trends towards next generation data centers have helped to drive our current research agenda. Our goal is to enable the flexibility and agility offered by these new technologies while enabling cost effective management, predictable behavior and improved quality of service. We will give an overview of our research on model driven design for enterprise application infrastructure, automated deployment, and operations of distributed application services executing in a virtualized, shared resource pool within these next generation data centers.Finally, we will summarize the implications, challenges, and opportunities posed by these trends on academic and industrial research. In particular, we consider the impact on software performance and software performance engineering and pose important unanswered questions.