Performance Update

1. Performance Update Eric L. Boyd Director of Performance Architecture and Technologies Internet2

2. Vision: Performance Information is … • Available • People can find it (Discovery) • “Community of trust” allows access across administrative domain boundaries (AA) • Ubiquitous • Widely deployed (Paths of interest covered) • Reliable (Consistently configured correctly) • Valuable • Actionable (Analysis suggests course of action) • Automatable (Applications act on data)

3. Getting There: Build & Empower the Community Decouple the Problem Space: • Analysis and Visualization • Performance Data Sharing • Performance Data Generation Grow the Footprint: • Clean APIs between each layer • Widespread deployment of measurement infrastructure • Widespread deployment of common performance measurement tools

4. Result: No more mystery … • Increase network awareness • Set user expectations accurately • Reduce diagnostic costs • Performance problems noticed early • Performance problems addressed efficiently • Network engineers can see & act outside their turf • Transform application design • Incorporate network intuition into application behavior

5. BWCTL (Bandwidth Controller) • What is it? A resource allocation and scheduling daemon for arbitration of iperf tests • Typical Solution • Run “iperf” or similar tool on two endpoints and hosts on intermediate paths • Typical road blocks • Need permissions on all systems involved • Need to coordinate testing with others • Need to run software on both sides with specified test parameters

6. BWCTL: 3-Party Flow Diagram

7. NDT: Network Diagnostic Tool • Web100 enhanced server handles testing and diagnostic services • Java based and command line clients allows testing from any client (local or remote) • Performance and configuration faults reported back to client • Drill-down functions provide more details & error reporting capabilities • Grant from NIH/NLM to explore duplex mismatch detection

8. Well Known NDT Server Web Request NDT - Server Client Redirect msg Web Browser Web Server Web Page Request Web page response Testing Engine Java Applet Test Request Control Channel Spawn child Child Test Engine Specific test channels NDT Flow Diagram

9. OWAMP: One-Way Active Measurement Protocol • What is it? • Measures one-way latency: 1-way ping • Control connection used to broker test request based upon policy restrictions and available resources. (Bandwidth/disk limits) • Specification • http://tools.ietf.org/wg/ippm/draft-ietf-ippm-owdp/draft-ietf-ippm-owdp-14.txt

10. OWAMP Flow Diagram

11. Thrulay Overview • Network capacity and delay tester • Same class of tools as iperf, netperf, nettest, nuttcp, ttcp, etc. • Unique features not found in other tools: • TCP: measures round-trip delay along with goodput • UDP: measures: • One-way delay, with quantiles • Packet loss • Packet duplication • Reordering • UDP: ability to send precisely positioned true Poisson streams (microsecond errors in sending times) • Human and machine-readable (ready to be fed to gnuplot)

12. Thrulay Update • New release v0.8 • Tests with multiple TCP streams • Set DSCP (a.k.a. first 6 bits of the TOS byte) • Report MTU and/or MSS (whichever the OS makes available) • More UDP statistics: duplication, reordering, quantiles of delay • SPARC/Solaris support • Mac OS X support • IPv6 support • Non-busy-waiting UDP mode (less precise, but can run more concurrent tests) • Documentation: manual pages have been added • Basic client authorization based on IP address • Integration of TSC timekeeping projects for faster and more precise timestamping

13. Bulk Transport • Build a library / tool for bulk transport that does not require kernel level modifications yet achieves the performance of such • VFER library • Congestion control hooks • Implements loss-based congestion control • Working on delay-based version • File transfer utility • An initial version demoed

14. How can you use them? • Tools are open source, supported, well-documented • BWCTL/Iperf, OWAMP, NDT are deployed across Abilene backbone and at many partners • You can: • See ongoing measurement results at the Abilene Observatory • Test to/from the Abilene backbone

15. Network Performance Measurement Workshops • Example Course Materials: • http://e2epi.internet2.edu/npw/presentations.html Goals: • Grow installed base of BWCTL/Iperf, OWAMP, and NDT at GigaPoP and regional campuses. • http://e2epi.internet2.edu/pipes/pmp/pmp-dir.html • Begin integration into IT support processes. • Create an installed base for perfSONAR deployment. • Give each participant tool-specific cookbooks.

16. Completed SOX / GaTech (03/05) CENIC / UCLA (06/05) JT – Vancouver (07/05) OARNet / OSU (09/05) MAGPI / FMM (09/05) MAX / College Park (12/05) APAN (01/06) JT - Albuquerque (02/06) MERIT (02/06) Columbia / NYSERNet (04/06) Planned University of Virginia (04/06) Under Consideration Wisconsin, Alaska, … Network Performance MeasurementWorkshop Locations and Dates

17. Additional Outreach • National Library of Medicine • Presentation on end-user tools at NLM-centric Network Performance Monitoring Pilot-project Workshop (March 2006) • New World Symphony • Presentation on end-user tools at NWS Master Class Workshop (January 2006) • On-going dialog on end-user needs

18. Collaboration • Working Groups • Bulk Transport (transport.internet2.edu) • GGF Network Measurement (nmwg.internet2.edu) • perfSONAR (www.perfsonar.net) • Funded Research • “Bridging the Gap” (NSF) • Network Measurement for International Connections • Google Summer of Code • 2005 very successful • 2006 planned

19. perfSONAR Overview • What: Measurement infrastructure for exchanging data under development • How: Webservices network performance framework • Network measurement tools • Network measurement archives • Distributed scheduling/authorization • Multi-domain policy • Common language (GGF NMWG Schema) • Where: Deployed / to be deployed across: • Network Backbones (Abilene, ESNet, GÉANT) • Regional Networks (NRENs, RONs, Gigapops) • Universities • When: First product release early summer ‘06

20. perfSONAR is a joint effort: ESnet Fermilab GÉANT2 JRA1 Internet2 RNP Internet2 includes: University of Delaware Georgia Tech Internet2 staff GÉANT2 JRA1 includes: Arnes Belnet Carnet Cesnet DANTE DFN FCCN GRNet ISTF PSNC Nordunet (Uninett) Renater RedIRIS Surfnet SWITCH perfSONAR Credits

21. How can you use it? • perfSONAR Link Utilization and Capacity data available from Abilene, ESnet, GÉANT (prototype) • Build your own components to integrate into open source framework

22. perfSONAR: Project Activity Meter • 1-2 conf calls/week • 1 new service/month (accelerating) • 3-4 development workshops/year • 3-4 paper submissions/year

23. perfSONAR: System Description • Domains represented by a set of services • Each domain can deploy services important to the domain • Analysis clients interact with service across multiple domains

24. perfSONAR: Services (1) • Lookup Service • Allows the client to discover the existing services and other LS services. • Dynamic: services registration themselves to the LS and mention their capabilities, they can also leave or be removed if a service gets down. • AuthN/Z Service • Internet2 MAT, GN2-JRA5 (eduGAIN) • Authorization functionality for the framework • Users can have several roles, the authorisation is done based on the user role. • Trust relationships defined between users affiliated with different administrative domains.

25. perfSONAR Services (2) • Transformation Service • Transform the data (aggregation, concatenation, correlation, translation, etc). • Topology Service • Make the network topology information available to the framework. • Find the closest MP, provide topology information for visualisation tools • Resource protector • Arbitrate the consumption of limited resources between multiple services.

26. Inter-domain perfSonar example interaction Useful graph Client Token MA Here is who I am, I’d like to access MA B Here is who I am, I’d like to access MA A Token MB a,b,c : Network A, MA A, AA A Where Link utilisation along - Path a,b,c? AA A Here you go Get Link utilisation a,b,c Get link utilisation c,d,e,f AA B Here you go a,b,c: Network A – LS A, c,d,e,f : Network B, MA B, AA B Where Link utilisation along - Path a,b,c,d,e,f? LS A LS B MA B MA A a b f e c d Network A Network B

27. perfSONAR: Status Update • Production release of base package expected by June (code freeze next week) • Will include: • Single domain LS solution • RRD MA • (no AS) • Additional services and client applications supporting this version will soon follow: • BWCTL MP • perfSONAR UI

28. perfSONAR: Hot Topics • Multi-domain hierarchical LS • AuthN/Z development plan with JRA-5 (eduGAIN) • SSH MP (LookingGlass) service • Topology Services • L2 specific MA service

29. MPs SSH/Telnet (Looking Glass) ABW (bandwidth packet capture cards) BWCTL NMS (SDH status) SNMP Command line (OWAMP, Ping, Traceroute) MAs RRD SQL TopS BWCTL Hades (owd, jitter, owpl) Flow replicator Visualization Clients CNM perfSONAR UI Visual perfsonar Looking glass perfSONAR: Current Developments

30. Questions?

31. perfSONAR: authN/Z plans • perfSONAR(JRA-1)/JRA-5 sub-group • Group tasked with determining how to leverage JRA-5 authentication system (eduGAIN) in perfSONAR infrastructure • Jeff Boote (Internet2) • Diego Lopez (RedIRIS) • Maurizio Molina (Dante) • Andreas Solberg (Uninett)

32. perfSONAR: Background • Designed with Federated authentication in mind • AS becomes a ‘proxy’ for Authorization requests

33. eduGAIN: Background • JRA-5 provided authentication “interface” • Provides “bridging” to other authentication systems • Shibboleth • PAPI • Others… • Designed mostly with web-browser interaction in mind

34. Current Status • Group has come to general consensus on how this should work • Paper is currently underway describing interaction of perfSONAR with eduGAIN API

35. perfSONAR: Trust relationship entities • Client • idP (identity provider) • pSR (perfSONAR resource “service”) • AS (perfSONAR AS service) • HLS (Home Location Service)

36. Automated Client Interaction

37. Normal User Interaction

38. Implications for JRA-5 • Future extensibility for multiple X.509 root CA certificates • Non-web profile for authN attribute request • Current identity provider servers (attribute stores) may need to hold attributes for non-human clients • Others???

39. Implications for JRA-1 • AS has slightly different role • Clients never directly interact with AS • AS is effectively a ‘proxy’ between services and the eduGAIN ‘bridging elements’ • Attribute requests from services to RP’s and from RP’s to AS need to be made in a ‘boolean’ fashion to protect the privacy of clients • Automated clients MUST have X.509 client certificates

40. Questions/Concerns • Let us know if you would like a copy of the ‘document’ when it is complete • Please feel free to send further questions/comments to Maurizio and Jeff boote@internet2.edu maurizio.molina@dante.org.uk

42. Deployment • Network Performance Workshops • Hands-on tool deployment training • GigaPoP/campus involvement • 250+ participants/11 Workshops • Global Deployments • Infrastructure (GÉANT2, RNP, Internet2, & ESnet) • Individual Tools (e2epi.internet2.edu/pipes/pmp/pmp-dir.html)

43. Performance Tools (1) • OWAMP: One-Way Active Measurement Protocol • One-way Delay • BWCTL: Bandwidth Control • Wraps NLANR / DAST-created Iperf • Measures throughput

44. Performance Tools (2) • NDT: Network Diagnostic Tool • Diagnoses common performance problems between a server and your laptop • Downloaded java applet from the web runs the tests • Thrulay • Network capacity and delay tester • TCP: measures round-trip delay along with goodput • UDP: measures: one-way delay, with quantiles; packet loss; packet duplication; reordering