On trains and wagons: switching variable length packets in slotted - PowerPoint PPT Presentation

On trains and wagons: switching variable length packets in slotted OPS Chris Develder Mario Pickavet Piet Demeester Dept. of Information Technology (INTEC) Ghent University - IMEC, Belgium UNIVERSITEIT GENT

Outline • Intro • Slotted variable length packets • Switch architecture • Performance criteria • Simulation set-up • Trains or wagons? - influence of load - influence of granularity - service differentiation • Conclusions COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 2

Optical switching • Optical switching: • direct light from an input port to an output port • possibly wavelength conversion • circuit-switching: • continuous bit-stream • pre-established light-paths • set-up: “manually” or automatic • packet/burst switching b c c • chunks of bits, encapsulated in packets b c a • packet header determines forwarding • e.g. label switching, GMPLS based d f f e f COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 3

Variable length packets in OPS • Segmentation & reassembly: - chop variable length packets into OPS slots - calls for extra S&R info in header 4 3 2 1 - S&R functionality resides at edges • “Trains or wagons”: - trains: treat train as a whole • S&R trivial since wagons are kept together 4 3 2 1 and in sequence • only a single header, i.e. minimal control overhead - wagons: treat each wagon individually 4 3 2 1 • simpler scheduling algorithms COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 4

Switch Architecture • Node in core OPS network (backbone) • Switch functionality: - slotted operation - WDM ports - fully non-blocking switching matrix (SOA based) - wavelength conversion to solve contention - FDLs to provide buffering F fibers all-optical W wavelengths space switch B buffer ports FDL delay = D COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 5

Scheduling • Scheduling: each timeslot: (0) collect packets (from inputs + FDLs) per destination output port (1) select packets for forwarding along outgoing fibres; (2) elect packets for buffering from excess packets; drop remaining packets 0 1 switch matrix 2 buffer drop COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 6

Simulation set-up • Parameters: - F=6 input/output fibres - W=8 wavelengths per i/o fibre - B=0..8 recirculating buffer ports - D=2L delay in buffer - L=1.5…20 wagons per train (average) • Traffic model: - train length: neg. exponential distribution rounded to slot length - train inter-arrival: Poisson process COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 7

Performance criteria • (byte) loss rate: - amount of data lost / amount of data sent - main indicator of service quality for end user • delay: - of secondary importance (delay in OPS switches only small fraction of end-to-end delay) • fairness: - large trains should not be discriminated against • service differentiation: - the scheduling mechanism should allow for efficient class of service differentiation COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 8

Influence of load (1) loss rate loss wagons / loss trains 1.E+00 200% 1.E-01 no buffer trains loss rate 1.E-02 better loss rate 1.E-03 100% B=4 B=8 1.E-04 wagons better 1.E-05 no buffer 0 0.2 0.4 0.6 0.8 1 4 buffer ports load 8 buffer ports trains, B=0 wagons, B=0 0% trains, B=4 wagons, B=4 0 0.2 0.4 0.6 0.8 1 trains, B=8 wagons, B=8 load • no buffer: trains better - wagon approach results in losing parts of multiple overlapping trains • with buffer: wagons can be better for medium loads - buffer allows to store wagons for multiple overlapping trains; wagon- approach allows to exploit buffer more efficiently than train approach COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 9

Influence of load (2) 0.10 total loss, 0.09 trains load=0.62 0.08 total loss, B=4 train loss rate 0.07 wagons 0.06 0.05 0.04 0.03 0.02 0.01 0.00 1 3 5 7 9 11 13 15 train size • fairness: - wagon approach seriously discriminates against longer trains • wagons can reach lower overall loss rate if sufficient buffer, and for medium load, but at the price of more unfairness (and possibly higher delays) COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 10

Influence of granularity 1.E-01 1.E-02 loss rate 1.E-03 wagons trains better B=4 better 1.E-04 0 5 10 15 20 25 meanpkt trains, load 0.5 w agons, load 0.5 trains, load 0.6 w agons, load 0.6 trains, load 0.7 w agons, load 0.7 • granularity: - performance of trains/wagons depends on ratio train length and OPS slot length • wagon approach better if trains are short (cross-over point shifts to slightly larger lengths for lower loads) COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 11

Service differentiation (1) • Scheduling: each timeslot: (0) collect packets (from inputs + FDLs) per destination output port (1) select packets for forwarding along outgoing fibres; (2) elect packets for buffering from excess packets; drop remaining packets ⇒ simple priority mechanism: first high priority packets 0 1 “priority queue”: 1) first higher priority packets; 2) same priority: first “oldest” 3) same timestamp: random switch matrix (uniform over same pri and tstamp) 2 tstamp = when packet enters switch buffer drop COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 12

Service differentiation (2) 1.E+00 loss rate 1.E-02 1.E-04 1.E-06 0.2 0.4 0.6 0.8 1 load high pri, trains high pri, wagons low pri, trains low pri, wagons • service differentiation: - train approach does not allow strong service differentiation with a simple differentiation mechanism without preemption of earlier arrived low priority trains • wagon approach achieves strong separation with very simple differentiation mechanism COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 13

Conclusions • wagon approach is advantageous… - …to achieve strong service differentiation - …to achieve lower overall loss for medium loads if there is a buffer - …to slightly reduce average delay when load is limited • … but at the price of - …stronger discrimination against long trains - …increased control overhead (header information + higher load on scheduler) COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 14

That’s all, folks! … thanks for your attention … any questions? UNIVERSITEIT GENT

Influence of load (3) delay delay wagons / delay trains 1.8 expo 1.6 160% expo 1.4 traindelay_recv 140% 1.2 1 120% 0.8 traindelay_recv 0.6 100% 0.4 80% 0.2 0 60% no buffer 0 0.2 0.4 0.6 0.8 1 40% load B=4 buffer ports trains, B=0 wagons, B=0 20% B=8 buffer ports trains, B=4 wagons, B=4 0% trains, B=8 wagons, B=8 0 0.2 0.4 0.6 0.8 1 load • low loads: wagon approach has slightly lower delays - only a few of the train’s wagons need to be buffered, whereas the train approach buffers complete trains (thus also the last wagon) • high loads: train approach has lower average delays - in wagon approach under high loads, the chance of having trains with no buffered wagons is substantially reduced COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 16

Delay measurement Tx 1 2 3 1 2 3 delay = 2 slots delay = 3 slots Rx 1 2 3 1 3 2 • delay: - delay induced by buffering - time elapsed between end of transmission of packet and completion of its reception • we account for possible re-ordering (with wagon approach) COIN, Tu.A2-6, 15 July 2003 C. Develder, et al., "On trains and wagons: switching variable packets in a slotted OPS network" 17