The Campus as key to Internet2 Engineering San Diego Guy Almes <almes@internet2.edu> 4 December 2000
Outline of Talk Internet2 Engineering Objectives The Logic of End-to-End Performance Our Aspirations Threats to these Aspirations Promising Approaches to Success The Internet2 End-to-End Performance Initiative
Internet2 Engineering Objectives Provide our universities with superlative networking: Performance Functionality Understanding Make superlative networking strategic for university research and education
The End to End Challenge Support advanced networking end to end Performance 100 Mb/s across the country normative several multiples possible in some cases Functionality Multicast Quality of Service IPv6 Measurements
What are our Aspirations? Candidate Answer #1: Switched 100BaseT + Well-provisioned Internet2 networking ® 80 Mb/s But user expectations and experiences vary widely
What are our Aspirations? Candidate Answer #2: Lower user expectations and minimize complaining phone calls There is a certain appeal I suppose...
What are our Aspirations? Candidate Answer #3: Raise expectations, encourage aggressive use, deliver on performance/functionality to key constituencies. Not the easy way, but necessary for success
Why should we Care? Advanced faculty needs: Effective access to remote facility: quickly move large datasets. PPDG: 400 Mb/s to CERN by 2003. Interactive access: video or control or VoIP. Very low loss/jitter. We (in several senses) need to deliver the goods.
Why should we Care? "We" as the university community. "We" as campus networking specialists. "We" as networking professionals. "We" as the (broad) Internet2 project. Low aspirations are dangerous to us.
Abilene core November 2000 Seattle New York Cleveland Indianapolis Sacramento Denver Washington Kansas City Los Angeles Atlanta Houston
Abilene Connections by (roughly) YE 2000
International Peering Seattle CA*net3, (AARnet) STAR TAP APAN/TransPAC, CA*net3, IUCC, NORDUnet, RENATER, REUNA, SURFnet, SingAREN, SINET, TAnet2 CERnet, (HARnet) OC12 New York DANTE*, JANET, NORDUnet, SURFnet CA*net3 Sunnyvale (SINET) Los Angeles SingAREN, SINET OC3-12 San Diego CUDI Miami (REUNA, RNP2, RETINA) El Paso (CUDI)
The Current Situation We have a combined Internet2 infrastructure of considerable capacity examples of 240 Mb/s flows End to end performance varies widely but 40 Mb/s flows not always predictable users don't know what their expectations should be Note the mismatch
The Current Situation We have a combined Internet2 infrastructure of considerable capacity examples of 240 Mb/s flows End to end performance varies widely but 40 Mb/s flows not always predictable users don't know what their expectations should be Note the mismatch
Threats to End to End Performance BW = C x packet-size / ( delay x sqrt(packet-loss )) (Mathis, Semke, Mahdavi, and Ott, CCR, July 1997) Context: Network capacity Geographical distance Aggressive application
Threats to End to End Performance Network Path local / department / campus gigaPoP / backbone / exchange points Host problems OS / TCP Hardware: NIC, CPU, memory, bus Application
Threats to End to End Performance Fiber problems dirty fiber dim lighting 'not quite right' connectors
Threats to End to End Performance Fiber problems Switches horsepower full vs half-duplex auto-sense 10/100 head-of-line blocking
Threats to End to End Performance Fiber problems Switches Inadvertently stingy provisioning mostly communication happens also in international settings
Threats to End to End Performance Fiber problems Switches Inadvertently stingy provisioning Wrong Routing asymmetric best use of Internet2 distance
Threats to End to End Performance Fiber problems Switches Inadvertently stingy provisioning Wrong Routing Host issues NIC OS / TCP stack CPU
Perverse Result 'Users' think the network is congested or that the Internet2 infrastructure cannot help them 'Planners' think the network is underutilized, no further investment needed, and users don't need high performance networks
Promising Approaches Work with key motivated users 'Shining a flashlight' on the problem Measurements Divide-and-Conquer Understanding Application Behavior Getting it right the first time
Active Measurements within Abilene Surveyors with: Active delay/loss measurements Ad hoc throughput tests
Application to Performance Debugging
Application to Performance Debugging
Divide and Conquer Systematically identify/isolate the network segment at fault Can we make this systematic and (eventually) automated?
End to End Advanced Functionality Multicast IPv6 QoS
Internet2 End-to-End Performance Initiative Distributed measurement infrastructure Teams of performance analysis specialists (PERTs) Dissemination of best practices
Defining End-to-End Success Metrics Identify core applications / services high-performance TCP VoIP / videoconferencing pervasive native IP multicast Scope How pervasive is it supported across the campus? Timeliness When are these metrics achieved?
Anticipated Partners NLANR Web100 Abilene partners Leading campuses and gigaPoPs Internet2 corporate partners
Initiative Phases 1st Gear 2nd Gear: Early Adopters Phase Preparation, planning, early experiments 2nd Gear: Early Adopters Phase Partner with 10-15 selected campus Develop PERTs, Measurement Infrastructure, etc. Build tools, resources, and best practices Expect RFP in late January 2001 3rd Gear: Dissemination Increasingly pervasive PERTs, infrastructure
Creating Internet2 Value Build the infrastructure together Make end-to-end performance and advanced functionality routine Identify and connect valuable resources for our faculty and students Have fun