The Internet & SLAC Les Cottrell1, SLAC http://www.slac.stanford.edu /grp/scs/net/talk/internet-connectivity-97/index.htm Outline – – – – – – of Talk I. SLAC’s connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 Talk presented at SLAC, July 1997 1 Some Acronyms ARA - Appletalk Remote Access, protocol to connect up remote Macs ATM - Autonomous Transfer Mode, a high high speed network mechanism DSL - Digital Subscriber Loop, a proposed medium speed (100s kbps - Mbps) leased line service (phone company answer to cable modems) ESnet - Energy Sciences network (DOE’s research network, SLAC’s main connection to Internet ISDN - Integrated Switched Digital Network, new <= 128 kbps digital switched phone service) POP - Point of Presence, a place where one or more networks have facilities SLIP/PPP protocols to provide Internet access over a serial line VPN - Virtual Private Network, a way of tunneling private data over the public Internet WAN - Wide Area Network 3/16/2016 2 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 3 Dial in Access http://www.slac.stanford.edu/comp/net/residential.html Terminal/emulator dial in – 7 ports, 14.4 kbps ARA – 16 ports, <=33.6kbps, ~340 accounts (85 active/mo) SLIP/PPP thru campus – 14.4kbps, need campus account Netcom, $15/mo, nationwide – 28.8 kbps Wireless via Ricochet ISDN Direct & via ISP – 9 ports, <=128 kbps, in pilot mode ~25 users, production service late summer Following VPN developments 3/16/2016 4 SLAC’s WAN Connectivity 43Mbps to ESnet ATM cloud (Sprint Oakland POP) 1.5Mps to Caltech/ESnet 1.5Mbps to LBNL/ESnet 10Mbps to Stanford 3/16/2016 5 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 6 What is Important to User We have to optimize the scarcest & therefore most valuable commodity - Time How long does it take after I hit the button? 3/16/2016 7 Value of Rapid Response Time Studies in late 70’s early 80s by Walt Doherty of IBM & others showed the economic value of rapid response time: – – – – – 0-0.4s 0.4-2s 2-12s 12s-600s >600s = High productivity interactive response = Fully interactive regime =Sporadically Interactive regime =Break in contact regime =Batch regime There is a threshold around 4-5s where complaints increase rapidly 3/16/2016 8 Ping Response for Groups of Hosts Average Ping Response for Various Groups of Hosts Seen from SLAC Jan-95 thru Dec-96 ESNET 700 Internatl Ping Response (ms) 600 NAmericaE NAmericaW 500 400 300 200 100 International little change N. America E improving 210 ms -> 150ms N. America W improving 140 ms -> 80ms 3/16/2016 ESnet improving 100ms -> 50 ms 6 -9 ov N -9 6 6 Se p ay M Ju l-9 -9 6 6 -9 ar M Ja n -9 6 5 -9 ov N Se p -9 5 5 Ju l-9 -9 5 M ay -9 ar M Ja n -9 5 5 0 9 European/Japan Packet Loss to SLAC Packet loss much more important – loss of packet typically causes 4-5s timeout Packet Loss to Major HEP International Sites seen from SLAC Jan-95 thru Apr-97 % 100 Byte Ping Packet Loss 45 40 CERN.CH Increase UK-US bandwidth DESY.DE IN2P3.FR RL.AC.UK KEK.JP ROMA1.INFN.IT 35 30 Improve Esnet Internet connect 25 20 15 10 5 0 Sep-94 Jan-95 Apr-95 Jul-95 Oct-95 Feb-96 May-96 Aug-96 Dec-96 3/16/2016 RAL: poor to unnacceptable, most others acceptable Mar-97 Jun-97 10 Quality by Host Group Ping Loss Quality Distributions for Host Groups <= 1% Loss (==Good) Percentile 70% (150, 0.79) (76, 5.46) (183, 7.18) >1% & <=5% Loss (==Acceptable) >5% & <=12% Loss (==Poor) > 12% & <=25% Loss (==Bad) >25% Loss (==Unusable) 60% 50% (188, 6.21) (199, 6.3) 40% 30% 20% 10% ic N Am er er N Am rn te In ic aW na at io ca l IS P Lo t Es ne (host-months, median loss) aE l 0% 0.0-1% Good, 1-5% Acceptable, 5-12% Poor 12-25% Bad, > 25% Unusable Similar to Internet Weather Report (<6%, <12%, > 12%) 3/16/2016 11 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 12 Driving Forces - Hosts 3/16/2016 13 Driving Forces - New Apps WWW, multimedia, Internet voice, video conferencing – > 60% graphics – <20% HTML 3/16/2016 14 Driving Forces - Penetration US Domains Countries with Internet access 3/16/2016 15 Current Internet Hosts 3/16/2016 16 Challenge - Diversity of Traffic other domains 13% it 2% Traffic out of FNAL ch 2% br 3% edu 43% gov 4% com 14% com 14% net 19% 3/16/2016 17 Challenge - No single Mgmt for Links 1 RTR-CGB4.SLAC.Stanford.EDU 2 RTR-DMZ.SLAC.Stanford.EDU 3 ESNET-A-GATEWAY.SLAC.Stanford.EDU 4 pppl-atms.es.net 5 nynap-pppl-atms.es.net 6 192.157.69.11 [Sprint NAP] 7 core3-hssi3-0.WestOrange.mci.net 8 core1.WestOrange.mci.net 9 border2-fddi-0.WestOrange.mci.net 10 border2-hssi1-0-gw.WestOrange.mci.net 11 192.204.183.3 [PREPnet] 12 DEFAULT1-GW.UPENN.EDU 13 NISC8.UPENN.EDU 3/16/2016 18 Challenge Commercial Internet focussed on staying alive as opposed to research or promoting advanced requirements 3/16/2016 19 More Acronymns CalREN2 - a California initiative to provide better educational & research networking CHEP97 - Computing in High Energy Physics meeting in Berlin, April 1997 ESSC - ESnet’s Steering Committee ICFA - International Committee on Future Accelerators Internet 2 - University initiative to provide improved networking between universities NGI - Next Generation Internet, Presidential initiative vBNS - very high-speed Backbone Network System, a high speed NSF funded backbone network 3/16/2016 20 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 21 New Initiatives - California CalREN2 – – – – 3/16/2016 joint proposal NSF, UC, Stanford, Caltech … includes Pac Bell & Cisco Distributed GigaPOPs in SF & LA, also SD & Sac Hi speed (622Mbps) ring around state envisioned 22 Bay Area Bay Area CalREN2 GigaPOP nodes: – UCSF – UCB (links to Esnet, Sprint, MCI/vBNS, UC Davis (state wide) – UCOP – Stanford (links to NASA/NSI, BBN, MCI & statewide ring) 3/16/2016 23 New U.S. Initiatives: vBNS NSF initiative for interconnecting supercomputer centers for “meritorious applications” Extended to promote University interconnectivity 622 Mbps backbone 3/16/2016 24 New US Initiatives: Internet 2 Started out (Oct-96) as consortium of ~ 34 major universities – Now there are over 100 covers 80% of US university sites we monitor – ~$500K / university over several years, 25% seed – Will use vBNS as backbone – GigaPOPs in major areas 3/16/2016 25 Next Generation Internet (NGI) Presidential Initiative – $100M/yr for 3 years – 100 sites at 100 times bandwidth (1.5Mbps => 155Mbps backbone) – 10 sites at 1000 times bandwidth – DARPA, DOE, NSF, NASA… Internet2/NGI/ESnet relationship unclear – can Universities connect to Internet 2 & ESnet? 3/16/2016 26 U.S. International Connections Only list those of interest to HEP Moving to colocate US end points at DC POP to improve peering Discussing CERN<=>Esnet<=>KEK link STAR-TAP = proposed Int’l GigaPOP at Chicago Country Brazil Canada CERN France Germany (DFN) Italy Japan Today (Jan-Mar '97) 128kbps to FNAL 2*45Mbps 2Mbps IN2P3 via CERN 1.54 Mbps to DC POP 1.54Mbps to PPPL KEK to FixW 512kbps UK (9 + 8.5)Mbps (ANS+Sprint) 45Mbps via Teleglobe (DC POP?) 3/16/2016 Plans Move to DC POP Move to DC POP May-97 ? Add part of 2*45Mbps Look to move to DC POP KEK 1.54Mbps to LBNL 27 Europe: TEN-34 W. European and some E. European countries interconnect at 4 - 34Mbps – de, it, ch, uk, gr, nl, pt, at, lu, es, fr, be, hu, sw+dk+no+fi – Several links in production, more by Jul-97 Intra country links generally good Intra Europe links improving with TEN-34 Next step TEN-155 3/16/2016 28 Asia & FSU Most connections thru Japan, in general good to acceptable for KEK – US/Esnet/KEK 522kbps => 1.5Mbps – China 64kbps => 128kbps (via KEK) – => 128kbps BINP/Russia Jun-97 2Mbps 3/16/2016 satellite DESY <=> MSU (Moscow) 29 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/ESnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 30 ESSC End-user Connectivity WG intra-ESnet connectivity good ESnet <=> University connectivity often bad ESSC set up WG to look at problem – – – – Ranked top 20 university sites by ER funding Monitored from SLAC & FNAL Identified worst (bad (6) to poor (8) performance) Recommended ESnet look at 6 BAD sites to understand costs of improving, DOE/ESnet will provide money – Typical Frame Relay 1.5Mbps connections $13K/month 3/16/2016 31 ESnet Peering Improved peering (63=>110 NSPs), examples: – MCI & Sprint to avoid public interconnect swamps – University of California (avoid Sprint) – THEnet at UT Austin – vBNS East Coast in place since Feb-97 (avoid W Orange MCI) West Coast May 1997 Chicago to come – Hubs at DC, Oakland, San Diego, Chicago – Now carry > 45K routes 3/16/2016 32 Improved ESnet Internet connection Cumulative % Packet Loss Weekday 7am - 7pm Packet Loss Washington.edu seen from SLAC '95 - '96 UOregon.edu UCSC.edu UCDavis.edu Colostate.edu Colorado.edu 100 90 80 70 60 50 40 30 20 10 S ep -9 6 N ov -9 6 6 Ju l-9 S ep -9 5 N ov -9 5 Ja n96 M ar -9 6 M ay -9 6 5 Ju l-9 Ja n9 5 M ar -9 5 M ay -9 5 0 ESnet Peers with Sprint/MCI to avoid MAE-West 3/16/2016 33 UC-Esnet Improved Peering & UCSC 25ms 3/16/2016 16ms 34 vBNS/Esnet Peering & U. Colorado Improved peering between Esnet & vBNS 3/16/2016 35 ICFA Internet Working Group Mini-workshop CHEP97 Working groups on: monitoring, remote regions, present status, requirements analysis, and the proposal End 1998 come up with proposal on what to do & why Next 3/16/2016 meeting Santa Fe, Sep-97 36 Monitoring - Why “You can’t manage what you can’t measure” Monitor to set “user” expectations, help with problem detection, get long term trends End-to-end monitoring mainly using ping Provides response time, packet loss, reachability, unpredictability Short (trouble shooting) & long term (planning) Most important metric is packet loss 3/16/2016 37 Monitoring - Who Many major HEP sites are monitoring endto-end Internet performance to collaborators – several hundred remote sites monitored Collaborative effort to provide HEP-wide and ESnet wide reports, requested by ICFA, ESnet – Partially funded by DOE FWP involving SLAC, LBL, HEPNRC – Based on SLAC early work (ping based) will complement LBL NIMI work – SLAC, HEPNRC/FNAL, LBL collaboration 3/16/2016 38 Monitoring - How Plan to coordinate effort, centered on SLAC/HEPNRC code – – – – install common software distributed architecture SLAC, HEPNRC Analysis Sites Umd, RAL, INFN, KEK, ARM, CMU, RMKI, IN2P3, CERN, DESY, TRIUMF, MSU signed up to be Collection Sites – 247 Remote Sites as of 7/7/97 – Reduces network impact of full mesh monitoring 3/16/2016 39 Data Collection & Distribution Architecture HTTP WWW E.g. HEPNRC E.g. SLAC Analysis Analysis E.g. RAL Collecting Collecting Ping Data (via HTTP) Collecting Collecting Pings Remote Remote Remote Remote Remote 3/16/2016 40 Results from ~70 Sites in 10 Countries Being Monitored from SLAC FNAL UMd SLAC ORNL Monitoring Site ESnet Site N. American Site International Site 3/16/2016 41 Putting it all Together 3/16/2016 42 Outline I. SLAC’s Connectivity II. How is it Working? III. Why is it like it is? IV. What’s going on out there on the Internet? V. What are we (DOE/Esnet, HEP, SLAC) doing? VII. Summary & Future 3/16/2016 43 Summary Driving forces: – Internet user growth 8.4M => 28M US users (15 mos) – Computer power doubling every 12-18 months – new applications, WWW, Internet phone, VR, Video ... Since Apr-95, no single management for planning, trouble reporting etc. ESnet performance good to acceptable, N. America poor (~6% packet loss avg), International poor (~7% packet loss avg) Bottlenecks at interchanges 3/16/2016 44 Future Many separate initiatives: – critical to make sure they interplay well – identify and avoid bottlenecks – understand and guide impact for HEP Criticality of Internet to HEP collaborations means HEP should increase efforts in this area: – keep tuned in, understand issues – monitor end-to-end performance – work with other research and higher education users 3/16/2016 45 SLAC http://www.slac.stanford.edu/comp/net/net.html SLAC Networking WAN Monitoring Page, lots of pointers http://www.slac.stanford.edu/comp/net/wan-mon.html ESnet: http://www.es.net/ vBNS: http://www.vbns.net/ Internet2: http://www.internet2.edu/ NGI: http://www.hpcc.gov/ngi-concept-08Apr97/ TEN-34: http://www.scimitar.terena.nl/projects/ten-34/ ICFA Workshop 3/16/2016 on HEP & the Internet: http://www.slac.stanford.edu/xorg/icfa/chep97/wg.html 46