E N D
Comparison of end-to-end packet ordering issues in synchronous and asynchronous Optical Packet Switching networksParticipants: Polytechnic University of Cartagena (UPCT), Polytechnic University of Catalonia (UPC)Responsible person:UPCT: Pablo Pavón-Mariño (pablo.pavon@upct.es), Joan Garcia-Haro (joang.haro@upct.es)UPC: Davide Careglio (careglio@ac.upc.edu), Miroslaw Klinkowski (mklinkow@ac.upc.edu)
OPS OPS OPS OPS OPS OPS OPS IP (electr.) IP (electr.) OPS WDM Backbone network
Tx E/O Tx E/O Tx E/O Previous work: synchronous OPS SCWP operational mode: Each node is free to choose transmission λ in ecah hop Header electronic processing 0 0 λ0 λ1 N-1 λ2 N-1 λ3 kin Optical input traffic kout Optical switching: tens of ns Optical buffering Optical switch fabric O/E IP sources (ingress traffic) Egress traffic Packet length ~ 1 μs
Previous work: synchronous OPS Round-Robin (proposed) WASPNET λ0 p9 p5 p1 p13 p11 p8 p4 p3 p1 λ0 p13 λ1 p10 p6 p2 λ1 p12 p9 p5 p2 λ2 p11 p7 p3 λ2 p10 p6 λ3 p12 p8 p4 λ3 p7 (b) Tx E/O 1 s Electronic memory P. Pavón Mariño, F. J. González-Castaño, J. García Haro, “Round-robin wavelength assignment: a new packet sequence criterion in optical packet switching SCWP networks”, accepted for publication in the European Transactions on Telecommunications, vol. 17, no. 4, Sept/Oct 2006.
Previous work: synchronous OPS KEOPS • OPS fabrics able to emulate output buffering + synchronization stage • Round-Robin ordering criteria network • Optimum scheduler proposed 0 0 0 0 0,..., n-1 n-1 0 1 n-1 • KEOPS, 2 I/O fibers, 32 λ, =80% • Bernoulli: B=2 => PLP<10-8 • ON-OFF (β=16): • B=2 (200m) => PLP=0.04 • B=10 (9 km) => PLP~10-5 N-1 0 0,..., n-1 (M-1) N-1 n-1
Synchronization stage Synchr. stage λ0 λ1 λ2 Advantages: Improves contention resolution => lower buffer requirements and delay Disadvantages: Cost
Our goals • Propose a round-robin criteria for ordering in asynchronous networks (done). • Propose an scheduler for output buffered architectures (like KEOPS) with no synch stages that preserves packet sequence following the new criteria (done). • Compare the performance of synch /asynch approaches, and try to answer the question: when and how the synch stages pay? (“under construction”) • A previous UPCT provides a markovian analysis for synch approach, under generical DBMAP traffic.