Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
The Earliest Deadline First Scheduling for Real-Time Traffic in the Internet Danny H. K. Tsang, Xiaojun Hei July 10, 2001 http://www.ee.ust.hk/~eetsang/ Department of Electrical and Electronic Engineering Hong Kong University of Science and Technology Outline • • • • Introduction Hierarchical scheduling framework EDF with active buffer management Conclusion and future work Broadband Network Lab 2 Network Convergence • Telephone Networks • Computer Networks • Cable Television Networks • Wireless Networks • Convergence: IP-centric networks Broadband Network Lab 3 Next Generation Internet • Broadband networks (WDM, 3G, 802.11, …) • Multimedia communication • Quality of Service provision – Integrated Services (IntServ) – Differentiated Services (DiffServ) Broadband Network Lab 4 Packet Scheduling • Service Policies • • First Come First Serve Generalized Processor Sharing • Earliest Deadline First • Buffer Management • • Drop-Tail Random Early Detection Broadband Network Lab 5 Real Time Traffic • Requirements: – High bandwidth – Time-sensitive – Packet loss • Transport candidates – User Datagram Protocol (UDP)--aggressive – Transmission Control Protocol (TCP)--cooperative Broadband Network Lab 6 Hierarchical Generalized Processor Sharing Link GPS GP S … GP S Class 1 Class N Broadband Network Lab 7 Hierarchical Scheduling Framework Link GPS Class 1 Class M GP S ……… … Best-effort Traffic EDF … Figure 1 Broadband Network Lab Real-time traffic 8 Hierarchical Scheduling Structure Link PGPS EDF CHOKe CHOKe P GP S Best-effort Traffic Real-time Traffic Figure 2 Broadband Network Lab 9 Queueing Model for the EDF-CHOKe Scheme flow 1 C CHOKe EDF … flow 2 flow N Broadband Network Lab 10 Dropping Probability (pa) RED 1 Average Queue length 0 min_th max_th Broadband Network Lab 11 CHOKe s tart Y AvgQsize<=M in _ t h ? N Draw a packet at random from queue Admit the new packet Y end Drop both packets Both packets from s ame flow? Y N AvgQsize<=M ax _ th? N Admit packets with a probability p Drop the new packet end end end Broadband Network Lab 12 Network Topology 1 1 10 M bp s UDP TCP Sink s p b 3 R2 … M 10 1 Mbps s bp M 33 R1 10 … TCP 3 2 10 s bp M 2 UDP Sink 33 Figure 3 Broadband Network Lab 13 UDP Throughput Comparison 5 10 x 10 FCFS-DropTail FCFS-RED FCFS-CHOKe EDF-CHOKe 9 8 Throughput(bps) 7 6 5 4 3 2 1 0 0 50 100 150 200 250 Time (sec) Figure 4 Broadband Network Lab 14 Per Flow Throughput Comparison 1000000 10000 Ideal 1000 FCFS-CHOKe EDF-CHOKe 100 10 31 28 25 22 19 16 13 10 7 4 1 1 Throughput(bps) 100000 Flow ID Figure 5 Broadband Network Lab 15 Fairness Index • Definition f ( x1 , x2 ,, xn ) 2 n ( xi ) i 1 n n xi i 1 2 Ideal 1.0 FCFS-DropTail 0.0305 FCFS-RED 0.0309 FCFS-CHOKe 0.3744 EDF-DropTail 0.0304 EDF-RED 0.0304 EDF-CHOKe 0.2838 Table 1 Broadband Network Lab 16 Packet Delay Distribution of the TCP Connections Figure 6 Broadband Network Lab 17 Statistics about the Delay Distribution(sec) TCP UDP Avg Std Avg Std FCFSCHOKe 0.9378 0.1203 0.0859 0.2743 EDFCHOKe 0.5665 0.0398 0.0476 0.1564 Table 2 Broadband Network Lab 18 Power of the Network Definition: throughput power delay TCP UDP FCFS-CHOKe 2.45 x 104 3.44 x 106 EDF-CHOKe 3.78 x 104 7.65 x 106 Table 3 Broadband Network Lab 19 Conclusion • EDF: Schedule real-time traffic • Hierarchical Scheduling Structure – Best-effort service together with real–time traffic • EDF with active buffer management – Good delay performance – Fair bandwidth allocation between UDP and TCP flows Broadband Network Lab 20 Future Work • Hierarchical scheduling framework • More complex networking scenarios Broadband Network Lab 21 Q&A Broadband Network Lab 22