Hedera_NSDI10_camera_ready

Hedera_NSDI10_camera_ready - Hedera: Dynamic Flow...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Hedera: Dynamic Flow Scheduling for Data Center Networks Mohammad Al-Fares * Sivasankar Radhakrishnan * Barath Raghavan † Nelson Huang * Amin Vahdat * * { malfares, sivasankar, nhuang, vahdat } @cs.ucsd.edu † [email protected] * Department of Computer Science and Engineering † Department of Computer Science University of California, San Diego Williams College Abstract Today’s data centers offer tremendous aggregate band- width to clusters of tens of thousands of machines. However, because of limited port densities in even the highest-end switches, data center topologies typically consist of multi-rooted trees with many equal-cost paths between any given pair of hosts. Existing IP multi- pathing protocols usually rely on per-flow static hashing and can cause substantial bandwidth losses due to long- term collisions. In this paper, we present Hedera, a scalable, dy- namic flow scheduling system that adaptively schedules a multi-stage switching fabric to efficiently utilize aggre- gate network resources. We describe our implementation using commodity switches and unmodified hosts, and show that for a simulated 8,192 host data center, Hedera delivers bisection bandwidth that is 96% of optimal and up to 113% better than static load-balancing methods. 1 Introduction At a rate and scale unforeseen just a few years ago, large organizations are building enormous data centers that support tens of thousands of machines; others are mov- ing their computation, storage, and operations to cloud- computing hosting providers. Many applications—from commodity application hosting to scientific computing to web search and MapReduce—require substantial intra- cluster bandwidth. As data centers and their applications continue to scale, scaling the capacity of the network fab- ric for potential all-to-all communication presents a par- ticular challenge. There are several properties of cloud-based applica- tions that make the problem of data center network de- sign difficult. First, data center workloads are a priori unknown to the network designer and will likely be vari- able over both time and space. As a result, static resource allocation is insufficient. Second, customers wish to run their software on commodity operating systems; there- fore, the network must deliver high bandwidth without requiring software or protocolchanges. Third, virtualiza- tion technology—commonly used by cloud-based host- ing providers to efficiently multiplex customers across physical machines—makes it difficult for customers to have guarantees that virtualized instances of applications run on the same physical rack. Without this physical lo- cality, applications face inter-rack network bottlenecks in traditional data center topologies [2]....
View Full Document

This note was uploaded on 11/12/2011 for the course CE 726 taught by Professor Staf during the Spring '11 term at SUNY Buffalo.

Page1 / 15

Hedera_NSDI10_camera_ready - Hedera: Dynamic Flow...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online