Energy Efficient PONs with Service Delay Guarantees and Long Reach

Transcription

1 Energy Efficient PONs with Service Delay Guarantees and Long Reach L. Valcarenghi, P. Castoldi TeCIP Institute Scuola Superiore Sant Anna, Pisa M. Chincoli CNIT, Pisa, Italy P. Monti and L. Wosinska ONLab, KTH, Kista, Sweden EE-LSDS Conference Apr Wien,

2 Summary Energy consumption in optical access networks overview Scenario PON Long Reach (LR) PON Cyclic sleep with Service based Variable sleep Period (CSVP) System description System model Results CSVP in Long-Reach PON Results Conclusions

3 Summary Energy consumption in optical access networks overview Scenario PON Long Reach (LR) PON Cyclic sleep with Service based Variable sleep Period (CSVP) System description System model Results CSVP in Long-Reach PON Results Conclusions

4 Energy Consumption in Communications Networks Access networks + mobile radio networks (except home networks) major contributors to energy consumption in communications networks: high number of Customer Premises Equipment (CPE) bandwidth underutilization Remaining part home networks optical access network energy consumption 50%-80% f wired networks energy consumption Source: C. Lange, D. Kosiankowski, R. Weidmann, and A. Gladisch, Energy Consumption of telecommunication networks and related improvement options, IEEE JSTQE, March/April, 2011 Access=Fixed Access Network= Fiber to the Exchange (FTTE) and Fiber to the Cabinet (FTTC): fiber + xdsl; FTTH (PON)

5 Considered Scenario PON and LR PON ONU-1 Central Office OLT Passive Splitter ONU-2 Idle traffic, but active receivers and service lines ONU-N Differential reaches ONU k LR-PON Up to 100 km and beyond ONU 1 ONU 2

6 Long Reach (LR) PONs Site reduction implies that longer distances must be covered by the same ODN Reach up to 100 km Longer distance -> longer RTT Node consolidation decreases energy consumption by decreasing the number of central offices Node consolidation increases energy consumption because of presence of amplifiers

7 Summary Energy consumption in optical access networks overview Scenario PON Long Reach (LR) PON Cyclic sleep with Service based Variable sleep Period (CSVP) System description System model Results CSVP in Long-Reach PON Results Conclusions

8 Service based variable sleep period Two main decisions must be made When to sleep For how long Reduce ONU energy consumption By means of service-based variable sleep period initially proposed by R. Kubo, et. al, Adaptive Power Saving Mechanism for 10 Gigabit Class PON Systems, IEICE Transactions on Communications, vol. 2, no. E93.B, pp , 2010 Guarantee delay constraints to applications subscribed by By computing the maximum allowed sleep time based on subscribed services and queuing theory system model End-user multimedia QoS categories, ITU-T Recommendation G.1010, nov Network performance objectives for IP-based services, ITU-T Recommendation Y.1541, Dec Ethernet frame transfer and availability performance, ITU-T IPTD=IP Recommendation packet Transfer Y.1563, Delay; Jan. IPDV=IP packet Delay Variation

9 Cyclic Sleep with Service-based Variable Sleep Period (CSVP) Rationale The OLT has a table in which services are categorized in terms of QoS class with The specific OLT is costantly bandwidth aware and of delay the services subscribed constraintsto each ONU The OLT builds a table containing the IPTD, IPDV, and minimum bandwidth requirement per each Each registered ONU is ONU subscribed to a list of services For each QoS class a specific sleep period is assigned to maximize the energy efficiency while Tsl guaranteeing is a function of the the constraints delay constraints 9

10 CSVP System description Modified SPW with service-based variable sleep period ONU Finite State Machine

11 CSVP: IP packet Transfer Delay (IPTD) model W q POLLING SYSTEM WITH GATED SERVICE POLICY WITH IDENTICAL STATIONS = V 2 V 2 V 2 Hp : Constant V + V V 2 ( ) 2 1+ λ S λ S ( ) + 1 λ S 2 ( 1 λ S ) = T sl + T OH + RTT Assumption: 1 ONU and 1 OLT W q Service Subscription Low load condition V 2 = T sl +T Update IPTD (min IPTD) λs << 1, λs + RTT 2 OH max IPTD = Wq + S + RTT / 2 2 << 1 W q delay V Queuing Vacation time=idle time Inform ONU Tsl = 2*(IPTDmax S RTT) S - TOH Service time Tsl

12 CSVP: IP packet Delay Variation (IPDV) model upper = quantile of the IPTD IPTDmin = minimum IPTD IPTDupper IPTDmax = TDS,max+RTT+Tsl+ +TOH+S+RTT/2 Frame arriving right after the CONFIRMATION Hp: low load condition λs << 1, TDS, max IPDV = IPTDupper- IPTDmin IPDV = TDS,max+RTT+Tsl+T OH 0 ITU-T Y.1541 def. IPDV = RTT+Tsl+TOH Assumption: 1 ONU and 1 OLT IPTDmin=S+RTT/2 Frame arriving right before the CONFIRMATION and no frame in the OLT queue Tsl = min{tsliptd,tslipd V} Service Subscription Update IPDV (min IPDV) Tsl = IPDV - TOH - RTT Inform ONU Tsl 12

13 Simulation Scenario CSVP Clas s ID Service characteristics Frame Payload size [Byte] Data rate Bi [b/s] IPTD [ms] IPD V [ms ] k k k k Simulation parameters 1 OLT, 1 ONU Event driven simulation Different QoS classes mix Negative exponential frame interarrival time distribution Parameter Variable Value Overhead time TOH 2ms Round-trip time RTT 0.4ms Power consumption in sleep mode Psl 1 W Power consumption in active mode Pa 10 W Channel rate C 10 Gb/s Number of span n 8

14 Results IPTD constraint guaranteed Tsl as a function of IPTD only IPDV computation based on the cumulative distribution function (CDF) of IPTD, uniformly distributed

15 Results IPTD and IPDV constraint guaranteed Tsl as a function of IPTD and IPDV IPDV computation based on the cumulative distribution function (CDF) of IPTD, uniformly distributed

16 Summary Energy consumption in optical access networks overview Scenario PON Long Reach (LR) PON Cyclic sleep with Service based Variable sleep Period (CSVP) System description System model Results CSVP in Long-Reach PON Results Conclusions

17 LR PON - Simulation Scenario CSVP Clas s ID Service characteristics Frame Payload size [Byte] Data rate Bi [b/s] Wq, a [ms] Wq, b [ms ] k U k k k Simulation parameters Event driven simulation Different QoS class mix Negative exponential frame interarrival time distribution Parameter Variable Value Overhead time TOH 0-2 ms Reach R km Power consumption in sleep mode Psl 1 W Power consumption in active mode Pa 10 W Channel rate C 10 Gb/s Number of span n 8

18 LR PON - Results Stricter delay constraint -> Reach heavily affects energy efficiency Increase in overhead time (i.e., an increased synchronization time) is more detrimental for energy efficiency than a reach increase Assumption: 1 ONU and 1 OLT TOH = 2ms Tsl = 2*(IPTDmax S RTT) - TOH

19 LR PON - Results (2) eta [%] 68% 66% 64% 62% 60% 58% Tc=2 ms, TOH=0.5 ms With fixed cycle time and higher number of ONUs, energy efficiently apparently increases because the slot for an ONU decreases but most of the time spent in scynchronization Contemporarily the bandwidth utilization per cycle decreases (smaller slots) ' E E P N 1 T # of ONUs eta [%] η = 68% 67% 66% 65% 64% 63% 62% 61% 60% 59% E = 1 eta P # of ONUs sl a Bw per ONU N OH T C 1.00E E E E+06 Bw per ONU

20 Conclusions Energy consumption in optical access networks Cyclic sleep with service-based variable sleep period (CSVP) Energy consumption reduction while providing delay guarantees Queuing theory model to compute sleep time Both average frame delay and frame jitter QoS constraints are met with different energy efficiencies Reach might heavily impact energy savings An increase in the number of ONUs decreases the energy consumption per ONU but at the expenses of a decreased bandwidth share, higher loss rate and delay per cycle

21 thank you! and thanks to