Analytical Approach to Dynamic Bandwidth Allocation Algorithm

used in LRPONMS Thesis Defense + PhD Qualifiers

Anu Mercian

Committee Members:

Martin Reisslein (Chair)

Michael McGarry

Cihan Tepedelenlioglu

Yanchao Zhang

Agenda

• Passive Optical Networks (PON)• Long-Range PON (LRPON)• Dynamic Bandwidth Allocation (DBA)• Part 1: Multi-threading Polling• Part 2: Parallel Polling• Part 3: Multi-polling techniques• Conclusion • Future Works

INTRODUCTION• We live in a bandwidth-hungry world with services

demanding superior performance in voice, data and video services

Passive Optical Networks• When the physical layer is Optic Fibers, bandwidth

obtained in the higher layers is promising

© wikipedia

Structure of PONDownstream Traffic

UpstreamTraffic

© [20] G. Kramer, G. Pesavento. Ethernet Passive Optical Network(EPON): Building a next Generation Optical Access Network

Evolution of PON• APON – ATM passive optical networks• BPON – Broadband PON (~622Mbps)• EPON – Ethernet PON (more widespread(easy

overlay))• GPON – higher bandwidth (larger variable length

packets) (ITU G.984)• GEPON – within the Ethernet, gigabit availability• LRPON – long-reach to allow >20km (~100km)

© Glen Kramer, Gerry Pesavento; Ethernet Passive Optical Network(EPON): Building a Next Generation Opitcal Access Network

Importance of LRPON• High CapEx and OpEx involved in PON deployment to cover

large area with a OLT/Central office for every 20Km of distribution

• LRPON covers 100Km reducing CapEx and OpEx• Reduction in active sites leads to larger distances

Multi Point Control Protocol

IEEE 802.3ah standard

© Michael P.McGarry, Martin Reisslein, Martin Maier; Ethernet Passive Optical Network Architectures and Dynamic Bandwidth Allocation Algorithms

Dynamic bandwidth allocation• Static Bandwidth Allocation – Each link is given a standard

BW allocated. • Disadvantages of SBA • Necessity of Dynamic Bandwidth Allocation – Statistical

Multiplexing• Design Space [22]:

– Grant scheduling framework – Grant sizing schemes – Scheduling Policies

© [22] M.P. McGarry; M. Reisslein. Investigation of the DBA Algorithm Design Space for EPONs

Grant Scheduling framework• Online or IPACT(Interleaved Polling with Adaptive

Cycle time)

• Offline

Grant sizing schemes• Fixed

• Gated

• Limited

Scheduling Policies• Shortest Propagation Delay

• Shortest Grant or Shortest Processing Time First(SPT)

Where τ is the half RTT

Problem Statement I• LRPON promises less OpEx and CapEx but the large

propagation delay gives poor delay performance. • A Solution – Multi-thread polling (MTP)[16]. • Discrepancies in MTP: MTP gives good delay

performance when compared to offline scheduling framework. But is it better than online technique?

© [16] H. Song et. al. Multi-thread polling: A Dynamic Bandwidth Distribution Scheme in LPON

Delay Analysis• Polling delay • Granting delay • Queuing Delay

• Channel Utilization[22]

Multi-Thread Polling• Idea is to send request before the previous Gate

message is received creating a new thread

© [16] H. Song et. al. Multi-thread polling: A Dynamic Bandwidth Distribution Scheme in LPON

Features of MTP• Tuning Multiple Threads

• Inter-thread Scheduling• Achieving Fairness

Reduced Delay in MTP• Polling delay

• Granting Delay

• Queuing delay

MTP: Analysis• MTP based on design space is (multiple-offline,

excess)• Polling delay for MTP is less because it has another

opportunity in the same cycle

• Granting delay for IPACT is less because IPACT is online, and does not involve wait time

• Overall delay for IPACT is lower

Experiment Settings• Simulator used: Simulator developed using CSIM

discrete event simulation library• Channel settings

– Channel Capacity C = 1 Gbps– Number of ONUs M = 16– Max Grant size = 7688bytes

• Self-similar traffic– Quad model packet size distribution– 60% 64bytes, 4% 300bytes, 11% 580bytes, 25% 1518bytes

• Same distance between ONU and OLT

ResultsDistance between ONU and OLT = 100Km

Overall Delay

Results

Channel Utilization Overall Delay

Distance between ONU and OLT = 20 Km

Problem Statement II• Offline technique was extended to obtain a multi-

thread process which gave delay performance better than offline but not as good as IPACT. What if online was extended as a multi-process?

Concept of Parallel Polling• Multiple online processes in parallel• Motivation:

– IPACT performs better than Multiple-offline threads so multi-online could be even better

– Issues in Multi-thread process: Problem of wait time or idle time

– Fairness Issue– Void formation

Parallel Polling: A possible solution• Each cycle time has two threads in parallel with each

one services as online

Advantages of PP• PP is based on online grant scheduling framework,

therefore low idle time and hence high utilization• No issue of fairness issue as GRANT is given as per

REPORTed• Thread tuning will be required and compared

between the effective cycle load• Easy and simple implementation

Delay performance of PP• Polling delay

• Granting delay

• Queuing delay

Delay Comparison• Polling delay of PP will be slightly less than MTP as

the cycle length of online process is less than the offline

• Granting delay for PP is very less when compared to MTP because of online technique

• Queuing delay is same for same simulator settings• Therefore,

ResultsDistance between ONU and OLT = 100Km

Overall DelayChannel Utilization

ResultsDistance between ONU and OLT = 20Km

Channel Utilization Overall Delay

Problem Statement III• MTP is better than offline scheduling framework• PP is better than online scheduling framework• So multiple polling of OLT is advantageous for

LRPON?

Multiple polling techniques• Multi-thread polling[16]• Parallel Polling• Double Phase polling[2], can be also called Multi-

group polling

© [2] S.Y. Choi et. al. Double Phase Polling Algorithm Based on Partitioned ONU Subgroups for High Utilization in EPONs

ResultsDistance between ONU and OLT = 100Km

Channel Utilization Overall Delay

ResultsDistance between ONU and OLT = 20Km

Overall DelayChannel Utilization

Conclusions• MTP is good for LRPON when compared to offline

but not when compared to online • PP gives comparatively best delay performance for

LRPON• Multiple polling techniques in one cycle time gives

promising results for LRPON with QoS awareness• Channel utilization of PP and IPACT are high when

compared to Multi-group, offline and MTP

Future Work• Can scheduling policies be useful for techniques

based on offline grant scheduling framework? • Can fairness be achieved in PP using online excess

bandwidth distribution (OEBD)• PP a new technique and can be explored further. will

it be promising for GPON as well?• Can multiple polling techniques be of use to LRPON• Consideration of performance of Real-time polling

with respect to current multi-polling techniques

• How will these DBA schemes perform with video and voice traffic.

• What if distances between ONU and OLT are random, how will the techniques perform?

• How can delay performance for PP be improved for SRPON.

THANK YOU

and

Questions?

References• [1] A.H. Lashkari; H.R. Zeidanldo; A.A.Sabeeh. Static Bandwidth Allocation on Optical Networks. In Proceedings of International

Conference on Machine Learning and Compu ting(IPCSIT), pages 498–503, Singapore, May 2011.• [2] S.Y. Choi; S. Lee; T.-J. Lee; M.Y. Chung; H. Choo. Double-Phase Polling Algorithm Based on Partitioned ONU Subgroups for

High Utilization in EPONs. IEEE Optical Communication Networks, 1(5):484–497, October 2009.• [3] A. Dixit; G. Das; B. Lannoo; D. Colle; M. Pickavet; P. Demeester. Adaptive Multi-Gate Polling with Void Filling for Long-

Reach Passive Optical Networks. In Proceedings of ICTON, pages 1–4, March 2011.• [4] A. Djupsjobacka. Time division multiplexing using optical switches. IEEE Journal on Selected Areas in Communications,

6(7):1227–1231, August 1988.• [5] O. Duffy. Poisson Traffic Model and Self similar traffic. In Proceedings of online website OwenDuffy.net, wikipedia, pages 1–4,

March 2010.• [6] M. Kamran. A framework for dynamic bandwidth allocation algorithms in TDM ethernet passive optical networks. In

Proceedings of International Symposium on High Capacity Optical Networks and Enabling Technologies, pages 1–5, October 2007.• [7] H. Song; B.-W. Kim; and B. Mukherjee. Long-Reach Optical Access Network: A Survey of Research Chanllenges,

Demonstrations, and Bandwidth Assignment Mechanisms. IEEE Communications Surveys & Tutorials, 12(1):112–123, June 2010.• [8] J.R. Kiniry. Wavelength division multiplexing: ultra high speed fiber optics. IEEE Internet Computing, 2(2):13–15, April 1998.• [9] M. Maier. WDM Passive Optical Networks and Beyond: the Road Ahead. IEEE/OSA Journal of Optical Communications and

Networking, 1(4):C1–C16, 2009.• [10] M.P. McGarry; M. Reisslein; M. Maier. Ethernet Passive Optical Network Architectures and Dynamic Bandwidth Allocation

Algorithms. IEEE Communications Surveys, 10(3):46–60, July 2008.• [11] A. Helmy; H. Fathallah; H. Mouftah. Interleaved Polling Versus Multi-Thread Polling for Bandwidth Allocation in Long-Reach

PONs. IEEE Optical Communications Networks, 4(3):210–218, March• 2012.• .

• [12] B. Kantarci; H. Mouftah. Bandwidth distribution Solutions for Performance Enhancement in Long-Reach Passive Optical Networks. IEEE Communications Surveys & Tutorials, PP(99):1–20, 2011.

• [13] B. Skubic; J. Chen; J. Ahmed; B. Chen; L. Wosinska; B. Mukherjee. Dynamic Bandwidth Allocation for Long-Reach PON: Overcoming Performance Degradation. IEEE Communications Magazine, pages 100–108, November 2010.

• [14] G. Kramer; B. Mukherjee. IPACT: A Dynamic Protocol for an Ethernet PON(EPON). IEEE Communications Magazine, pages 74–80, February 2002.

• [15] H. Song; A. Banerjee; B.-W. Kim; B. Mukherjee. Multi-Thread Polling: A Dynamic Bandwidth Distribution Scheme in Long-Reach PON. In Proceedings of IEEE GLOBECOM, pages 2450–2454, March 2007.

• [16] H. Song; B.-W. Kim; B. Mukherjee. Multi-Thread Polling: A Dynamic Bandwidth Distribution Scheme in Long-Reach PON. IEEE Journal on Selected Areas in Communication, 27(2):134–142, February 2009.

• [17] L. Shi; S. S. Lee; H. Song; B. Mukherjee. Energy-Efficient Long-Reach Passive Optical Network: A Network Planning Approach Based on User Behaviors. IEEE Systems Journal, 4(4):449–457, December 2010.

• [18] F.J. Effenberger; K. McCammon; V. O’Byrne. Passive optical network deployment in North America. Journal of Optical Networking, 6(7):808–818, July 2007.

• [19] G. Kramer; B. Mukherjee; G. Pesavento. Ethernet PON(ePON): Design and Analysis of an Optical Access Network. Photonic Network Communications, 3(3):307–319, August 2001.

• [20] G. Kramer; G. Pesavento. Ethernet Passive Optical Network(EPON): Building a Next Generation Opitcal Access Network. IEEE Communications Magazine, pages 66–73, February 2002.

• 47[21] J.R. Ferguson; M. Reisslein; and M.P. McGarry. Online excess bandwidth distribution for Ethernet passive optical networks. IEEE Optical Networking, 8(4):358–369, April 2009.

• [22] M.P. McGarry; M. Reisslein. Investigation of the DBA Algorithm Design Space for EPONs. IEEE/OSA Journal of Lightwave Technology, 30(14):2271–2280, 2012.

• [23] N.A.M. Radzi; N.Md. Din; M.H. Al-Mansoori; I.S. Mustafa; S. Kh; Sadon. Efficient Dynamic Bandwidth Allocation Algorithm for Upstream EPON. In Proceedings of IEEE 9th Internation Conference on Communications, pages 376–380, Kuala Lumpur, Malaysia, December 2009.

• [24] M.P. McGarry; M. Reisslein; C.J. Colbourn; M. Maier; F. Aurzada; M. Scheutziw. Just-in-Time Scheduling for Multi-channel EPONs. IEEE Systems Journal, 26(10):1204–1216, May 2008.

• [25] X. Bai; C. Assi; A. Shami. On the fairness of dynamic bandwidth allocation schemes in Ethernet Passive Optical Networks. Computer Communications, 29(11):2125–2135, 2006.

• [26] C. Wu; X. Shi; X. Yang; J. Su. The Impact of Parallel and Multithread Mechanism on Network Processor Performance. In Proceedings of the Fifth International Conference on Grid and Cooperative Computing, pages 1–5, October 2006.

• [27] W.-P. Chen; W.-F. Wang; W.-S.Hwang. Adaptive dynamic bandwidth allocation algorithm with sorting report messages for Ethernet passive optical network. IET Communications, 4(18):2230–2239, 2010

BACK UP

Basis study• Availability of many DBA schemes in research brings

a need to classify these DBA schemes based on their importance

Classification Description• Existent classifications cover QoS aware and unaware

but their usability is not explicitly mentioned• Connecting scheduling framework to user traffic type

and demand– Direct– Predictive– Intelligent

EPON-LRPON Classification

Usability• Direct – User’s that require high throughput but do

not compensate on fairness• Predictive – User’s that have a flexible requirement• Intelligent – When user’s of high load and low load

are combined

Results: Poisson

Channel Utilization Overall Delay

20Km

100Km

Results: Poisson

20Km

100Km

Channel Utilization Overall Delay

Results: Poisson

20Km

100Km

Channel Utilization Overall Delay

Design Space Equations• Offline

• Online

• DPP

• MTP

• PP

Idle time Equations

Multi-thread Polling

Parallel Polling