This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Optimizing a Virtualized Data Center David Erickson, Brandon Heller, Shuang Yang, Jonathan Chu, Jonathan Ellithorpe, Nick McKeown, Guru Parulkar, Mendel Rosenblum Stanford University Scott Whyte, Stephen Stuart Google ABSTRACT Many data centers extensively use virtual machines (VMs), which provide the flexibility to move workload among physi- cal servers. VMs can be placed to maximize application per- formance, power efficiency, or even fault tolerance. However, VMs are typically repositioned without considering network topology, congestion, or traffic routes. In this demo, we show a system, Virtue, which enables the comparison of different algorithms for VM placement and network routing at the scale of an entire data center. Our goal is to understand how placement and routing affect over- all application performance by varying the types and mix of workloads, network topologies, and compute resources; these parameters will be available for demo attendees to explore. Categories and Subject Descriptors: C.2.3 [Computer- Communication Networks]: Network Operations General Terms: Experimentation, Measurement, Performance Keywords: Data center network, OpenFlow, Virtualization, Virtue 1. INTRODUCTION Over the past decade, virtual machines have grown in pop- ularity as application containers, since they improve server utilization and support fast provisioning for scale-out ser- vices. Increasingly, VMs are hosted in clouds, large data centers that host thousands of customer VMs. Amazon EC2, the leading cloud provider, has approximately 40,000 servers , launches 80,000 VMs each day, and has launched 23 million VMs since its inception . Equipment and energy costs provide a strong motivation for cloud owners to maximize operational efficiency. One way to improve operational efficiency is through workload placement algorithms, which map VMs onto physical ma- chines (PMs). A placement algorithm might squeeze VMs onto as few servers as possible, then power down the un- needed servers or sell access to them on a spot market. Al- ternately, it might spread the VMs as evenly as possible, to maximize headroom for new VMs or avoid bandwidth bot- tlenecks. The placement algorithm could even maximize the performance of individual services, by co-locating their VMs or considering policy requests from customers. Currently published workload placement algorithms en- Copyright is held by the author/owner(s). SIGCOMM11, August 1519, 2011, Toronto, Ontario, Canada. ACM 978-1-4503-0797-0/11/08. Experiment Generator Initial Routing/ VM Placement Workload Runner (DNRC) Data Center Simulator (Local Machine) Optimized Routing/ VM Placement or HW Sim Measured Statistics Virtue Figure 1: Experimenters workflow using Virtue force CPU, RAM, and NIC sharing policies [8, 9]; notably absent from this list is the network fabric interconnecting...
View Full Document
This note was uploaded on 12/01/2011 for the course EE 5373 taught by Professor Chao during the Spring '11 term at NYU Poly.
- Spring '11