------------------------------------------- Minutes for the Ad-hoc CCIRN QoS Meeting ------------------------------------------- 26th August 1998 Sheraton Hotel, Chicago, USA Attendees: Guy Almes Internet2 almes@internet2.edu John Coulter CANARIE (CA) jcoulter@on.bell.ca Luca Dell'Agnello INFN-GARR (IT) luca@cnaf.infn.it Javad Boroumand NSF (US) jborouma@nsf.gov Franca Fiumana INFN-GARR (IT) fiumana@cnaf.infn.it Ruediger Geib Deutsche Telekom geib@advanced.org Shigeki Goto Waseda Univ (JP) goto@goto.info.waseda.ac.jp Yasuichi Kitamura CRL (APAN) kita@crl.go.jp Simon Leinen SWITCH (CH) simon@switch.ch Paul Love Internet2 epl@internet2.edu Kevin Meynell (Chair) TERENA K.Meynell@terena.nl Vishy Narayan NASA-NREN vnarayan@mail.arc.nasa.gov Niels den Otter SURFnet (NL) otter@surfnet.nl Trond Skjesol Uninett (NO) Trond@uninett.no Chuck Song MCI csong@mci.net Ben Teitelbaum (Sec) Internet2 ben@internet2.edu Paul Zawada NCSA/StarTAP zawada@ncsa.uiuc.edu ------------- DANTE TEN-34 Kevin Meynell for Peter Gietz (DANTE) ------------- Started in 1996 DANTE = 16 European research nets 50% funded by EU Included a program to test advanced networking issues Runs over ATM bearer service provided by PTOs, not told whose switches. TF-TEN conducted various experiments across JAMES - TEN-34 connects to telco ATM service - experiments with CBR, VBR PVCs - also tested SVCs - limited testing of ABR tunneled through CBRs - worked well but far from production quality - RSVP testing between Switzerland, Netherlands, Italy, France - no major problems with small scale - Problems with: - SVCs over the infrastructure - finding apps to test the net TEN-34 finished at end of July; net will continue until Dec JAMES finished in March QUANTUM (QUALity Network Tech for User-oriented Multimedia) proposal submitted to 5th call of EU Fourth Framework Programme TEN-155 Status - production ATM over STM-1 - single provider - managed bw for a # of services (IP, VPN research, MBONE) - Growing body of opinion in Europe that ATM is not way to go. But this goes with he conservative approach and considers it the only mech that is capable of fulfilling these reqs now. TF-TEN - testing program (probably jointly done by DANTE and TERENA) - RSVP - IP DiffServ - ATM SVCs with QoS - traffic measurement - hi-fi videoconferencing Tend to have one POP/country - historical for Europe. SDH can now be purchased in many parts, but not all, of Europe. URLs - TEN-34 (http://www.dante.org.uk/ten-34/) - TF-TEN (http://www.dante.org.uk/ten-34/tf-ten/) - QUANTUM (http://www.dante.org.uk/ten-34/quantum/) - ABR test results (http://www.fys.ruu.nl/~atm/documents/tm) Discussion: Chuck S.: "What does the "155" refer to?" A: Country PoP inter-connect. Chuck S.: "Does TEN-155 provide US connectivity" A: Not now - individual countries (NRNs) do that; focus is on intra-european. TEN-34 DOES include a US connection. connectivity. Javad: Is SDH available in Europe? A*: Regulatory problems. Question about DANTE QoS on US link. A: Only contractual, but many countries exceed contract. ---------- MCI/vBNS Chuck Song ---------- vBNS Arch and Requirements IP over ATM - LIS and UBR PVC mesh (for BE) - QoS on the same IP service - IP is primary Layered QoS architecture vBNS Topology map - pops and SCs - about 60 university connections - FORE switches (vBNS) and Newbridge (commercial Hyperstream switches) High Speed Direct User Connections - OC12 backbone - OC12/OC3 customer ports - large BW flows and reservation requirements - requirement of high-performance for QoS mechanisms Nationwide Backbone Service - individual ports capable of filling backbone - # flows << than on commercial backbone - currently 4 SCCs, 4NAPs, and 50 universities (out-of-date (May)) - limited number of flows Traffic Characteristics - high bw, very bursty vBNS QoS Design Rationale - useful service - similar to CL - must support OC3 level of BW commitment Service Requirements - high-performance for OC12 packet rates - initially, not addressing full scalability - not many connected sits, but most of these are capable of dumping data at rate to fill the backbone. - must extend to DS for scalability as vBNS adds more nodes vBNS Reserved BW - vBNS runs an ATM service on top of another ATM service (MCI's Hyperstream commercial service) - measurability (bw w/ bursts, time window in secs, shaping not required) - invocation (dynamic with RSVP) - granularity (from per-app, to pre-port) - scope (vBNS edge-to-edge) - unicast initially (only SharedExplicit style RSVP) - bw, not delay stronger than CL service - not GS service, shaping not available everywhere ATM COnfiguration vBNS Switch - co-lo with Hyperstream switches - Q.2931 signalling for VBR SVC - UBR mesh for RSVP Path and Resv messages Hyperstream Switch - concatenation of hop-by-hop VBR PVPs - UBR VPs are not IP over ATM - signal mapping - UBR PVCs for Path and Resv messages - CL Rspec mapped to VBR UPC - SVC call triggered by Resv packet coming back from next-hop router - SVC uses concatenated VBR VP's - route chosen by PNNI on vBNS - RT delay before setup complete Router Enhancement - joint MCI/Cisco project - Cisco (design and implementation) - vBNS (testing) Router Packet Classification - dynamically set up by RSVP - flow granularity: ACL, header fields, port, etc. - flow identity used for selection of QoS VCC - potential extension to DS with various config and setup utilities Router Packet Policing - actively or passively monitor class bits - average and peak rates, and burst size flow Queueing - per-flow queueing - "out" packets on same queue and "in" - misordering avoided - dynamic sharing of common buffer pool - min: no starvation of individual flows - max: no monopoly of all buffer sace Packet Discard - WRED - statistically penalizes "bad" data sources - per-class: min, max, drop prob - like RIO Issues - which ATM VCC for signalling packets - measurement of QoS traffic perf - Multicast, MARS, based or other - ADmissions control - need IP/ATM co-ordination - potential congestion in both routers and switches Policy Server DiffServ Trials - initially EF PHB - static configuration - before bw broker available - provisioning difficulty avoided first - expect DS traffic to shape - users = Internet2 community Status - testing Reserved BW Discussion: (no post talk discussion; some comments integrated into the notes above) ------------ CA*Net2 John Coulter ------------ initial plan: 2 class of services with parallel nets funding ends March 1999 initial hope to interoperate with vBNS May not be able to do much with vBNS as CA2 is ending Mar99 and the vBNS is just starting - a bit later than first thought. therefore, plan is now to try out bits of DS CA*Net3 Overview - IP over WDM - 8 wavelengths per route - 4 reserved for traditional SONET 4/BLSR by carrier - initial operation 25 Sept - nationwide by July 1999 - Can't break open the Chicago piece until BellCanada has carrier status in the US, so link to StarTap will stay ATM from Toronto. - First link, Montreal-Ottawa-Toronto target is 25 Sept (2 weeks before OECD) Lots of questions about IP-over-WDM - use of SONET but for framing and regen only; no ADMs; simple Nortel regens - the future: WDM directly to the router QoS - focus on QoS is at the IP layer - working closely with Internet2 ------------------ APAN Update Shigeki Goto for Dae Young KIM (unable to attend) APAN WG for QoS/CoS (chair is Professor Dae Young KIM from Korea) ------------------- - discussing intserv and diffserv issues - high bw US-Japan link about to be operational - testing over Japan-Korea and Japan-Singapore link - not compatible with TEN-34 -------------- NASA NREN QoS Vishy Narayan -------------- All links OC3 - Hope to be site for NGIX West - All sites FORE switches with Cisco routers behind - connections to Sprint switches - policing done at Sprint COs - 5 main sites Status Five month ago - "QoS!! What QoS?" Status Now - ATM QoS - testing wide-area ATM service conformance - aides in transitioning services/apps over to NISN - Wide Are Tests - UBR/CBR - video tests - network parameters measured and tuned for optimum performance - Currently full-mesh; Future plans to move to PNNI - Wide-are tests (CBR) - GSFC to Ames - use of ATM conversion device to transmit data RSVP and Multicast - between Goddard and Lewis Research Center - Test IP Multicast that makes use of RSVP - MC tools vic and vat used to transmit a source from one location to another - source stream is NASA TV broadcast video - medium-resolution frames (320x240) (15-30 fps) - real-time shuttle launch digital video distribution from NASA Kennedy Space Center - initial goal: 5-10 cameras - goal: 400 cameras - pre-launch video availability Cont... - GUI to pick views selectively - ATM and IP based QoS - PIM dense mode - extensive use of native IP multicast over NREN backbone ------------------------ Internet2 QoS, Diffserv, and the "QoSBone" Ben Teitelbaum ------------------------- --------------------- Concluding Discussion --------------------- Consensus that having informal informational exchanges like this meeting is productive. Perhaps meet like this ~6 months WG will meet again at Orlando IETF and San Jose INET meeting How can WG be more of a "working" group, rather than just a forum for information exchange? - could the "QoSBone" idea be internationalized? - difficulty of coordinating any kind of experiments over expensive trans-oceanic links - Guy: there would be tremendous value to coordinated (but not contiguous) QoS deployments/experiments - after the technologies are proven, can extend to int'l links, where the QoS is really needed Disseminate technically cooperation with the QoSBone to TERENA. Participation by international networks. Administrative stuff - Ben will send Kevin the minutes - All presenters will send Kevin their slides - Kevin will explode the minutes and MIME-attached slides to all participants.