Flow Routing to Power NGN IP Services. Dr. Lawrence G. Roberts CEO, Anagran

Transcription

1 Flow Routing to Power NGN IP Services Dr. Lawrence G. Roberts CEO, Anagran

2 The Beginning of the Internet - ARPANET 1965 MIT Two Computer Experiment Proved Circuit Switching too slow, unreliable and expensive Packet Network Proposed by Roberts Queuing Theory Len Kleinrock Packet Structure from Experiment Computer Interface from Experiment Roberts moved to ARPA ARPANET Planned All ARPA Research Sites pushed to join network 1968 RFP for Packet Switch won by BBN 1969 Steve Crocker led student team to design host-host protocol 1969 First 4 nodes installed: UCLA, SRI, UCSB, U. Utah 1. Toward a Cooperative Network of Time-Shared Computers,Roberts & Marill, Fall Joint Computer Conference, SF, California, November Multiple Computer Networks and Intercomputer Communication, Lawrence Roberts, ACM Symposium, Gatlinburg, Tennessee, October 1967.

3 Internet Early History 100,000 Internet Name first used- RFC 675 Roberts term at ARPA 10,000 Kahn term at ARPA Cerf term at ARPA SATNET - Satellite to UK Hosts or Traffic in bps/10 1, Aloha-Packet Radio PacketRadioNET Spans US Ethernet FTP TCP/IP Design NCP DNS TCP/IP Hosts Traffic 10 ICCC Demo 1 X.25 Virtual Circuit standard

4 NAE Draper Award Laureates Feb. 20th, 2001 Roberts Kahn Kleinrock Cerf

5 Planned Use of Internet Voice Totally moving to packets Low Loss required Video - Totally moving to packets Low Loss required Downloads Needs Faster File Transfer Streaming Needs Fast Guaranteed Rate Path Setup Emergency Services No Preference Currently Preference - Who gets service when capacity limited Authorization required

6 Changing Structure of Internet Was: Low Speed Edge, High speed Core No way to Overload the Core Unlimited use was OK Now: Broadband Edge, Core Limited Economically Edge Speed is for Burst Speed, not Continuous use Unlimited use not a reasonable option Edge Traffic must be controlled

7 Coming Soon A World Divided IPv4 IPv6 ASIA must convert to IPv6 No address space left Many Devices will not have IPv4 addresses Mobile, RFID USA will not convert as soon plenty of addresses left Many programs and devices will only speak IPv4 Dual Stack does not solve this users cannot interact IPv4-IPv6 Translation will be required Translation required many places, huge capacity

8 Change Is Required Address Space running out Need IPv6 No provision for Video Too many flows packet routers drop packets from all Cannot determine if path exists with low loss & jitter at rate Need Preference (who gets it) File Transfer much too slow and inefficient Cost growing as traffic grows Correction needed Traffic growing faster than Moore s Law - % GDP an issue Fairness required P2P can swamp other users Delay must be reduced for rapid response

9 A New Architecture Flow Routing

10 What is a Flow Router? An Individual Flow is a stream of packets between one user/system and another: web, voice, video, data In IPv4 it is uniquely identified by the 5- tupple (Dest. Address, Source Address, Protocol, Dest. Port, Source Port) In IPv6 it is uniquely identified by the 3-tuple (D-Address. S -Address, Flow Label) Composite Flows are groups of individual flows Similar to MPLS but with thousands of rate controlled classes Route the flows, not just packets Now rates exist - manage the rates BEFORE congestion hits Flow Routers keep track of state information for each individual flow and all composite flows

11 Evolution of Router Architectures 80 Gbps Routers Packet Routers Route every Packet WRED Major Packet Delay No Rate Control No Load Balance 8 Gbps per RU First Generation Flow Routers Route First Packet WFQ Major Packet Delay Flow Rate Control P2P rate control 8 Gbps per RU Second Generation Flow Routers Route First Packet Per Flow Policing Minor Packet Delay Flow & Group Rate Control P2P rate/route control 80 Gbps per RU

12 Packet and Flow Router Design Packet Router Route Big Output Buffer Switch Major Cost Major Delay Discard Output Input Load Controlled at Inputs Load Controlled at Outputs Flow Router Route Switch Lo a d M Flow State Discard easure ment Minor Delay Output Input Same Capacity at 1/5 the Power and Size

13 IPv6-IPv4 Translation = Flow State IPv4 Flow Router IPv6 Translation will be required throughout the network Major translation throughput will be required Translation hosts will be insufficient in capacity IPv4 to IPv6 translation requires flow state memory For the IPv6 address, ports, and flow label Flow Routers have the flow state to do the conversions Throughput per RU: Flow Router = 24 Gbps, Server = 1 Gbps Flow router can also do routing and P2P control

14 Overcapacity was to avoid Overload TCP Packet Routers Packet Routers under Overload Drop 30-50% of Traffic - Flows Synchronize Output Oscillates TCP Flow Routers Flow Routers under Overload Drop 1-3% of Traffic - Independent Discards avoid Synchronization Packet Router - WRED - 2:1 Overload Flow Router - IFD - 2:1 Overload 100% 75% 50% 25% 0% 100% 75% 50% 25% 520 0% 1, Utilization UDP Utilization Packet Routers Drop Packets All 6 TV Programs Ruined UDP Flow Routers Drop Flows Five TV Programs Perfect, One Dropped Overloaded Packet Routers destroy both TCP & UDP Overloaded Flow Routers maintain quality for both TCP & UDP

15 Flow Routers Identify & Control P2P UncontrolledP2P with WRED Rate VoD P2P controlled to fair rate Dimensions Rate Total Bytes Total Time Packet Size Port Protocol DiffServ P2P adjusted down if it continues HTTP VoIP Time Controls Rate ze Priority i Delay Var. ts e Loss Rate ck a P Problem: Classify flows so that each class of flow can be shaped for overall quality Deep Packet Inspection can do this at high expense but fails with encryption Flow Router rules watch the flow rate, byte count, and packet size over time This separates large long P2P, interactive, VoIP, etc. allowing separate shaping

16 Preference Priority Who gets the capacity? Preference Priority was in the telephone network It never has been in the Internet With fixed rate streams it is critical in overloads: Emergency Services and Military Priority for the office who gets through Priority for the home who gets the HDTV stream Flow Routers include preference priority For fixed rate flows, the lowest priority may get blocked For file transfer, the lowest priority may get a lower rate

17 Where? The Data Center Vastly Improve Performance, Scalability, & QoE Increase Utilization over 90% Load Balance over multiple paths Reduce packet loss for files, video, voice Reduce power requirements

18 Where? The ISP Aggregation Layer Flow Router Content Sources Text Images Video Voice Aggregation Layer CORE L2 Switches Flow Router WiFi or Mobile Array Rate control insures fairness, P2P and overload control Both on the outbound links (DSLAM) and for the individual users (DSL) High quality delivery of IPTV & VoIP; low delay, jitter & loss Multiple SLA s can be supported and reported DDoS attack identification and protection

19 Conclusions IPv6 use will grow due to address shortage Dual Stack insufficient, Translation required, Flow State needed The need for IPTV and P2P Control also require change Precedence now required for Emergency Services Flow Routers solve these problems: Insure Video and Voice quality by limiting the # of flows Reduce delay, jitter, and packet loss, increase utilization Precedence for emergency services, fixed rate streams Incorporate flow identification and rate control P2P controlled Reduce network cost and power over 5:1