PlanetLab: A Platform for Planetary-Scale Services Mic Bowman

Slides:



Advertisements
Similar presentations
PlanetLab: An Overlay Testbed for Broad-Coverage Services Bavier, Bowman, Chun, Culler, Peterson, Roscoe, Wawrzoniak Presented by Jason Waddle.
Advertisements

1 Planetary Network Testbed Larry Peterson Princeton University.
Chapter 4 Infrastructure as a Service (IaaS)
The future of Desktops Transform Your Desktop with Virtualization.
Xen , Linux Vserver , Planet Lab
INTRODUCTION TO CLOUD COMPUTING CS 595 LECTURE 6 2/13/2015.
Empowering Business in Real Time. © Copyright 2009, OSIsoft Inc. All rights Reserved. Virtualization and HA PI Systems: Three strategies to keep your PI.
PlanetLab: An open platform for developing, deploying, and accessing planetary-scale services Overview Adapted from Peterson.
Copyright 2009 FUJITSU TECHNOLOGY SOLUTIONS PRIMERGY Servers and Windows Server® 2008 R2 Benefit from an efficient, high performance and flexible platform.
Web Caching Schemes1 A Survey of Web Caching Schemes for the Internet Jia Wang.
1 GENI: Global Environment for Network Innovations Jennifer Rexford Princeton University
Microsoft Virtual Server 2005 Product Overview Mikael Nyström – TrueSec AB MVP Windows Server – Setup/Deployment Mikael Nyström – TrueSec AB MVP Windows.
1© Copyright 2015 EMC Corporation. All rights reserved. SDN INTELLIGENT NETWORKING IMPLICATIONS FOR END-TO-END INTERNETWORKING Simone Mangiante Senior.
Nikolay Tomitov Technical Trainer SoftAcad.bg.  What are Amazon Web services (AWS) ?  What’s cool when developing with AWS ?  Architecture of AWS 
The Future of the Internet Jennifer Rexford ’91 Computer Science Department Princeton University
Presented by Sujit Tilak. Evolution of Client/Server Architecture Clients & Server on different computer systems Local Area Network for Server and Client.
Bandwidth DoS Attacks and Defenses Robert Morris Frans Kaashoek, Hari Balakrishnan, Students MIT LCS.
Cloud computing Tahani aljehani.
Platform as a Service (PaaS)
Google AppEngine. Google App Engine enables you to build and host web apps on the same systems that power Google applications. App Engine offers fast.
CERN IT Department CH-1211 Genève 23 Switzerland t Next generation of virtual infrastructure with Hyper-V Michal Kwiatek, Juraj Sucik, Rafal.
1 Content Distribution Networks. 2 Replication Issues Request distribution: how to transparently distribute requests for content among replication servers.
Windows ® Powered NAS. Agenda Windows Powered NAS Windows Powered NAS Key Technologies in Windows Powered NAS Key Technologies in Windows Powered NAS.
Active Network Applications Tom Anderson University of Washington.
Adam Leidigh Brandon Pyle Bernardo Ruiz Daniel Nakamura Arianna Campos.
Software to Data model Lenos Vacanas, Stelios Sotiriadis, Euripides Petrakis Technical University of Crete (TUC), Greece Workshop.
Cloud Computing Saneel Bidaye uni-slb2181. What is Cloud Computing? Cloud Computing refers to both the applications delivered as services over the Internet.
Distributed Systems Concepts and Design Chapter 10: Peer-to-Peer Systems Bruce Hammer, Steve Wallis, Raymond Ho.
A Cloud is a type of parallel and distributed system consisting of a collection of inter- connected and virtualized computers that are dynamically provisioned.
SCAN: a Scalable, Adaptive, Secure and Network-aware Content Distribution Network Yan Chen CS Department Northwestern University.
VENUE: AFE BABALOLA HALL, UNIVERSITY OF LAGOS PRESENTATION BY: MR AMOS EMMANUEL CHIEF SOFTWARE ARCHITECT, PROGRAMOS SOFTWARE CHAIRMAN,
Benefits: Increased server utilization Reduced IT TCO Improved IT agility.
PlanetLab Applications and Federation Kiyohide NAKAUCHI NICT 23 rd ITRC Symposium 2008/05/16 Aki NAKAO Utokyo / NICT
Intel IT Overlay Jeff Sedayao PlanetLab Workshop at HPLABS May 11, 2006.
An Overview of the PlanetLab SeungHo Lee.
By L. Peterson, Princeton T.Anderson, UW D. Culler, T. Roscoe, Intel, Berkeley HotNets-I (Infrastructure panel), 2002 Presenter Shobana Padmanabhan Discussion.
1 A Blueprint for Introducing Disruptive Technology into the Internet Larry Peterson Princeton University / Intel Research.
Using Virtual Servers for the CERN Windows infrastructure Emmanuel Ormancey, Alberto Pace CERN, Information Technology Department.
Advanced Computer Networks Topic 2: Characterization of Distributed Systems.
Kiew-Hong Chua a.k.a Francis Computer Network Presentation 12/5/00.
DISTRIBUTED COMPUTING Introduction Dr. Yingwu Zhu.
Distributed Computing Systems CSCI 4780/6780. Distributed System A distributed system is: A collection of independent computers that appears to its users.
Distributed Computing Systems CSCI 4780/6780. Geographical Scalability Challenges Synchronous communication –Waiting for a reply does not scale well!!
PROP: A Scalable and Reliable P2P Assisted Proxy Streaming System Computer Science Department College of William and Mary Lei Guo, Songqing Chen, and Xiaodong.
6/23/2005 R. GARDNER OSG Baseline Services 1 OSG Baseline Services In my talk I’d like to discuss two questions:  What capabilities are we aiming for.
 Cachet Technologies 1998 Cachet Technologies Technology Overview February 1998.
Plethora: Infrastructure and System Design. Introduction Peer-to-Peer (P2P) networks: –Self-organizing distributed systems –Nodes receive and provide.
Globus and PlanetLab Resource Management Solutions Compared M. Ripeanu, M. Bowman, J. Chase, I. Foster, M. Milenkovic Presented by Dionysis Logothetis.
Infrastructure for Data Warehouses. Basics Of Data Access Data Store Machine Memory Buffer Memory Cache Data Store Buffer Bus Structure.
Web Technologies Lecture 13 Introduction to cloud computing.
Picturex Secures and Scales Event-Photo Sharing for Enterprise and Private Customers by Relying on the Powerful, Scalable Microsoft Azure Platform MICROSOFT.
09/13/04 CDA 6506 Network Architecture and Client/Server Computing Peer-to-Peer Computing and Content Distribution Networks by Zornitza Genova Prodanoff.
TRUST Self-Organizing Systems Emin G ü n Sirer, Cornell University.
Michael Miller Senior Director Real-Time Collaboration Products Oracle Collaboration Suite 10g Oracle Corporation.
Hosting Wide-Area Network Testbeds: Policy Considerations Larry Peterson Princeton University.
1 A Blueprint for Introducing Disruptive Technology into the Internet Larry Peterson Princeton University / Intel Research.
Intro to Distributed Systems Hank Levy. 23/20/2016 Distributed Systems Nearly all systems today are distributed in some way, e.g.: –they use –they.
John S. Otto Mario A. Sánchez John P. Rula Fabián E. Bustamante Northwestern, EECS.
Virtual Server Server Self Service Center (S3C) JI July.
4a. Aula 2o. Período de Livro texto Copyright © 2012, Elsevier Inc. All rights reserved March 5, 2012 Prof. Kai Hwang, USC Cloud Roles in.
© 2007 IBM Corporation IBM Software Strategy Group IBM Google Announcement on Internet-Scale Computing (“Cloud Computing Model”) Oct 8, 2007 IBM Confidential.
01/27/10 What is PlanetLab? A planet-wide testbed for the R & D of network applications and distributed computing Over 1068 nodes at 493 sites, primarily.
Platform as a Service (PaaS)
Chapter 6: Securing the Cloud
Platform as a Service (PaaS)
Software Defined Networking (SDN)
AWS Cloud Computing Masaki.
Specialized Cloud Architectures
Towards Distributed Test-Lab for Planetary-Scale Services
EE 122: Lecture 22 (Overlay Networks)
Presentation transcript:

PlanetLab: A Platform for Planetary-Scale Services Mic Bowman

Agenda What Is PlanetLab? Planetary-Scale Services –Evolving the Internet Why PlanetLab?

PlanetLab Is… Technology: –An open, global network test-bed for inventing novel planetary-scale services. –A model for introducing innovations into the Internet through the use of overlay networks. Organization: –A collaborative effort involving academic and corporate researchers from around the world –Hosted by Princeton, Washington, Berkeley, and MIT; sponsored by Intel, HP, and Google Socially –Cutting edge research infrastructure made available to the global community

PlanetLab Is… IA32 servers (836  1000’s) connected to the Internet at 412 sites Federated with PlanetLab Europe Mostly standard Linux distribution and dev environment A few global services

Other brands and names are the property of their respective owners. Academic Partipants

Other brands and names are the property of their respective owners. Industry Participants

Agenda What Is PlanetLab? Planetary-Scale Services –Evolving the Internet Architecture Why PlanetLab?

Content Distribution, 1993 NCSA’s “What’s New” the most viewed page on the web (100K accesses per month). All clients access a single copy of the page stored on a single server. End-to-End design works pretty well for store-and-forward applications

Content Distribution, 1998 IBM web “server” handles a record 100K hits per minute at the Nagano Olympics DFS running on SP2’s used to distribute 70K pages to 9 geographically distributed locations End-to-End design breaks down at scale (flash crowds, global distribution, …)

Content Distribution Today A Planetary-Scale Service Edge services provide 1000’s of points of presence throughout the Internet Overlay networks are constructed to move the content around efficiently The transition from “end-to-end” to “overlay” enables reliable planetary-scale services

Planetary-Scale Services Pervasive –Runs everywhere, all the time Robust –Robust system from flaky components Adaptive –Aware of and adapts to changing environment Scalable –Scales to a global workload

To Build One, You Need… Multiple vantage points on the network –Near the edge—low latency to clients –Near the core—good connectivity –Global presence A little computation at many locations –Computation beyond a single machine –Computation beyond a single organization Management services appropriate to the task –Resource allocation –Provisioning and configuration –Monitoring nodes, services, networks But who can afford it? –No single app can justify the infrastructure costs –Network today is like big-iron before timeshare

Solution: Share the Platform Everyone contributes a piece of the platform; everyone can use the whole platform –Build a “time-sharing” network-service platform –Cost shared among all the apps using it Model of future public computing utility –Nodes owned by many organizations –Shared cooperatively to provide resilience Platform must provide –Isolation to protect services from one another –Market-based resource allocation

PlanetLab Service Architecture Node 1 Node 2 Node 3 Node 5 Node 4 Mgmt. VM Hardware VMM Service Virtual Machines

PlanetLab Services are Running EventProcessingNetworkMappingDistributed Hash Tables ContentDistribution Web Casting Infrastructure Services & End-user Services Node 1 Node 2 Node 3 Node 5 Node 4

Resource Reservations CPU resources can be scarce during certain periods (before paper deadlines) The Sirius Resource Calendar Service allows PlanetLab users to schedule an increase a slice’s CPU priority for certain time periods –Only CPU and not work Seems to work well: –Rarely 50% subscribed –Services often deal with CPU loading themselves

PlanetLab Today… 836 IA32 machines at 412 sites –Principally universities, some enterprise –Research networks: I2, CANet/4, RNP, CERNet –Globally distributed –Some co-location centers –Federated with PlanetLab Europe Machines virtualized at syscall level –Name space isolation for security –Network, CPU, memory, file system isolation –Interface is a Linux machine with minimal install Complete access to the network

What We Got Right Immediate impact –Within 18 months 25% of publications at top OS & Comm conferences were PlanetLab experiments –Became a “expectation” for validation of large system results –And we learned some very interesting things

What We Got Right (continued) Incident response –Early: very conservative Don’t get turned off before value is established –Later: less restrictions Local administrators defend their researchers –Education Researchers: the kind of experiment that causes alarms Administrators: touchy IDS implementations

We Could Have Done Better Community contributions to the infrastructure –Infrastructure development remained centralized, we are paying the price now Support for long-running services –Researchers aren’t motivated to keep services running for multiple years –Decreased the amount of service composition (can’t trust the dependent services will continue to run)

We Could Have Done Better (continued) Admission control –Good practices make it possible to run many experiments, but very easy to consume all resources

Open Challenges Community ownership of availability –Need to motivate decentralized management Who keeps the nodes running? What happens when the nodes aren’t running? Resource allocation aligned objectives –Performance, innovation, stability

Open Challenges (continued) Standardization –Standard interfaces  platform stability –Open architecture  improved innovation Tech Transfer

Agenda What Is PlanetLab? Planetary-Scale Services –Evolving the Internet Architecture Why PlanetLab?

PlanetLab and Industry Global communications company –Incubator for future Internet infrastructure –Emerging services become a part of the Internet Global computer vendor –Platform for planetary-scale services –Need to understand for our customers Software company –Testbed for next generation applications –Cost-effective way to test new ideas Fortune 500 company –Next generation opportunities for IT staff –Leverage deployed PlanetLab services for CDN, object location, network health…

Summary PlanetLab is: –A globally distributed testbed that facilitates experimentation and deployment of scalable Internet services. The testbed has successfully established itself as a platform for cutting edge research. –Active research community using it for a wide variety of technologies. –Multiple papers published top academic conferences, e.g. OSDI, SOSP, NSDI, Sigcomm, … –300+ active projects Come join the fun (

BACKUP

Princeton: CoDeeN Content distribution –Partial replication of content –Redirect requests to optimal location of content PlanetLab Deployment –100 nodes, 150+ GB of data moved among the sites –Working to build service redirector Key Learnings –First service targeted for end users (proxy cache) –Maintaining server health is hard and unpredictable BBB B B B B A A A A AA C C C C CCC

UWashington: Scriptroute Distributed Internet debugging and measurement –Distribute measurement points throughout the network –Allow user to connect & make a measurement (upload scripts) PlanetLab Deployment –Running on about 100 nodes –Basic service used by other services Observations –Experiments look like port scan attacks –Low BW traffic to lots of addrs breaks some routers –Scriptroute adjusted spray of packets to avoid the problem

Cornell: Beehive DHT for object location –High performance –Self-organizing –Scalable Proactive-replication –Hash buckets replicated –O(1) lookup times for queries CoDoNs: DNS replacement –High performance P2P –Adaptive, load balancing –Cache coherent

Usage Stats Slices: 600+ Users: Bytes-per-day: 4 TB IP-flows-per-day: 190M Unique IP-addrs-per-day: 1M (source: Larry Peterson, May 2007)