Thesis Proposal draft Advisor Professor Frank Y S

  • Slides: 56
Download presentation
Thesis Proposal (draft) Advisor: Professor Frank Y. S. Lin Present by Yi-Wei Li Information

Thesis Proposal (draft) Advisor: Professor Frank Y. S. Lin Present by Yi-Wei Li Information Management Dept. National Taiwan University 1 2020/11/23

Title �基於配置頻寬、流量控制與路由決策達成具端點至 端點延遲及吞吐量公平之網路營運 �Joint flow control, capacity allocation and routing to achieve end-to-end delay

Title �基於配置頻寬、流量控制與路由決策達成具端點至 端點延遲及吞吐量公平之網路營運 �Joint flow control, capacity allocation and routing to achieve end-to-end delay and throughput fairness in wireless mesh networks. 2 2020/11/23

Outline �Introduction �Problem description �Problem formulation 3 2020/11/23

Outline �Introduction �Problem description �Problem formulation 3 2020/11/23

Introduction 4 2020/11/23

Introduction 4 2020/11/23

Background �WMN �Architecture �Advantage 5 2020/11/23

Background �WMN �Architecture �Advantage 5 2020/11/23

Delay vs. Throughput 6 The delay curve of the M/M/1 queueing model 2020/11/23

Delay vs. Throughput 6 The delay curve of the M/M/1 queueing model 2020/11/23

Motivation �Qo. S must be considered by communication protocols �End-to-end mean delay �Delay jitter

Motivation �Qo. S must be considered by communication protocols �End-to-end mean delay �Delay jitter �Packet loss ratios �Fairness �Aggregate and per-user throughput 7 2020/11/23

Motivation �To ensure reliable and high-quality network services, routing and capacity assignment policies should

Motivation �To ensure reliable and high-quality network services, routing and capacity assignment policies should be carefully designed. �In addition, we also consider flow control. 8 Network Planning Performance Assurance & Optimization Network Monitoring Traffic Modeling Network Capacity Expansion Network Servicing 2020/11/23

Fairness �Max-min fairness �Maximizing the network use allocated to the sessions with the minimum

Fairness �Max-min fairness �Maximizing the network use allocated to the sessions with the minimum allocation �Proportional fairness �Trying to maximize total wireless network throughput while at the same time allowing all users at least a minimal level of service. �Balanced fairness • For all pairs of flow a, b, the relative change in the capacity allocated to a when b is removed is the same as the relative change in the capacity allocated to b when a is removed. 9 2020/11/23

Problem description 10 2020/11/23

Problem description 10 2020/11/23

Scenario … … 11 2020/11/23

Scenario … … 11 2020/11/23

Scenario … … 12 2020/11/23

Scenario … … 12 2020/11/23

Scenario … … 13 2020/11/23

Scenario … … 13 2020/11/23

Problem description �Given a network topology, how to achieve 100% fair performance �It is

Problem description �Given a network topology, how to achieve 100% fair performance �It is an interesting problem when we have to decide both a routing path and input traffic rate for each O-D pair to maximize the smallest input rate and fairness �The delay close form of M/M/1 queue: � It is easy to show that it is a nonconvex programming problem by verifying the Hessian of when Cl and fl are both to decision variables 14 2020/11/23

Assumption �We assume our system can arbitrarily divide 15 capacity of air interface in

Assumption �We assume our system can arbitrarily divide 15 capacity of air interface in routers into any size related to link l. �All of network nodes in our system communicate with each other in a contention-free environment. �We use Kleinrock’s independent assumption that there is a buffer at each outbound link and this buffer is modeled as an M/M/1 queue. �The buffer we discussed here has infinite size and retransmission is not considered in our model. �There is a function in mesh routers that we can measure mean delay associated with flow ingress to that node. 2020/11/23

Given �Network topology �Capacity of air interface in each router �Position of mesh clients

Given �Network topology �Capacity of air interface in each router �Position of mesh clients �Traffic requirement of O-D pairs �Maximum link utility that aggregated flow would not exceed �With gateway functionality associated with capacity Cg, mesh routers can be connected to the Internet 16 2020/11/23

To determine �Routing path of each O-D pair � Fairness �Input rate of each

To determine �Routing path of each O-D pair � Fairness �Input rate of each O-D pair �End-to-end delay of each O-D pair 17 2020/11/23

Objective and constraints �If all clients have no requirements, we can achieve 100% fairness.

Objective and constraints �If all clients have no requirements, we can achieve 100% fairness. �In order to avoid this situation, we would like to maximize the minimum amount of allowed input. �Subject to: �Qo. S constraints �Flow constraints �Routing constraints 18 2020/11/23

Problem formulation 19 2020/11/23

Problem formulation 19 2020/11/23

Given parameters Notation Description N The set of network nodes which include mesh routers

Given parameters Notation Description N The set of network nodes which include mesh routers and mesh clients L The set of communication links in the communication network Ln The set of links whose roots are node n N W The set of O-D pairs in the network Cn (pkt/sec) The capacity of each air interface at network node n N 20 2020/11/23

Given parameters (Cont. ) Notation Description Pw A given set of simple directed paths

Given parameters (Cont. ) Notation Description Pw A given set of simple directed paths from the original to the destination of O-D pair w The indicator function which is 1 if link is on path and 0 otherwise (pkt/sec) The arrival rate of new traffic for each O -D pair w W, which is modeled as a Poisson process for illustration purpose A given link utilization factor which keeps link from saturated 21 2020/11/23

Decision variables Notation Description (pkt/sec) The allowed arrival rate of new traffic for each

Decision variables Notation Description (pkt/sec) The allowed arrival rate of new traffic for each O-D pair w W cl (pkt/sec) The capacity of each link l L The maximum allowable end-to-end delay for O-D pair w 22 yw (pkt/sec) The rejected arrival rate of new traffic for each O-D pair w W xp A routing decision variable which is 1 when path p Pw is used to transmit the packets for O-D pair w and 0 otherwise gl The aggregate flow over link l, which equals to Dw The end-to-end delay for O-D pair w W 2020/11/23

Objective function: 23 2020/11/23

Objective function: 23 2020/11/23

Extending work �Dimension 1: Uplink and downlink 24 2020/11/23

Extending work �Dimension 1: Uplink and downlink 24 2020/11/23

Thanks for your listening! 25 2020/11/23

Thanks for your listening! 25 2020/11/23

Compare with previous research Our work Previous work Routing Flow control 26 2020/11/23

Compare with previous research Our work Previous work Routing Flow control 26 2020/11/23

27 2020/11/23

27 2020/11/23

Problem description �有一些fairness研究採用routing 及capacity assignment來達成fairness(舉出文獻) �由於我們想在受限於per-user之capacity constraints,在提過的(100% fairness)演算法下, 最佳化系統效能。 �從這兩個觀點切入很難達到各自最佳化,因此我們 想用integer programming來分析彼此間tradeoff (fairness V.

Problem description �有一些fairness研究採用routing 及capacity assignment來達成fairness(舉出文獻) �由於我們想在受限於per-user之capacity constraints,在提過的(100% fairness)演算法下, 最佳化系統效能。 �從這兩個觀點切入很難達到各自最佳化,因此我們 想用integer programming來分析彼此間tradeoff (fairness V. S. 系統效能) �我們除了上敘兩變因,仍加入了flow control �Multicast tree (tree v. s. core node) �Reverse-multicast tree (Steiner tree) 28 2020/11/23

Objective function 29 2020/11/23

Objective function 29 2020/11/23

Subject to: = 0 or 1 30 2020/11/23

Subject to: = 0 or 1 30 2020/11/23

Hard fairness �情境2 31 2020/11/23

Hard fairness �情境2 31 2020/11/23

Upstream � Assumption 1. Router 可控管其下所屬頻寬且可任意切割為最小單位之集 合 2. Contention-free environment 3. 網路中拓樸及設備已固定(router,gateway,channel assignment) 4.

Upstream � Assumption 1. Router 可控管其下所屬頻寬且可任意切割為最小單位之集 合 2. Contention-free environment 3. 網路中拓樸及設備已固定(router,gateway,channel assignment) 4. 使用single path routing 5. processing delay、transmission delay、propagation delay皆為定值 6. 系統內節點以 M/M/1 queue表示 32 2020/11/23

Upstream Gateway Internet 1 2 3 a 4 6 33 5 b c Transient

Upstream Gateway Internet 1 2 3 a 4 6 33 5 b c Transient access point (TAP) Mobile client 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 34 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 34 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 35 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 35 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 36 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 36 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 37 b c 2020/11/23

Gateway Internet 1 2 4 3 5 6 a 37 b c 2020/11/23

Downstream 38 2020/11/23

Downstream 38 2020/11/23

39 2020/11/23

39 2020/11/23

40 解C 8, C 9 2020/11/23

40 解C 8, C 9 2020/11/23

由上解得知 解C 6, C 7 41 2020/11/23

由上解得知 解C 6, C 7 41 2020/11/23

由上解得知 解C 3, C 4, C 5 42 2020/11/23

由上解得知 解C 3, C 4, C 5 42 2020/11/23

由上解得知 解C 1, C 2 43 2020/11/23

由上解得知 解C 1, C 2 43 2020/11/23

Max-min capacity allocation (model 2) �Assumption �任意切割頻寬 �Contention free �Tree-based �Given: �To be determine:

Max-min capacity allocation (model 2) �Assumption �任意切割頻寬 �Contention free �Tree-based �Given: �To be determine: Cl , Xp ( , ), 44 2020/11/23

Max-min capacity allocation (model 2) �Objective: max min �Subject to �Flow <= Capacity �Per

Max-min capacity allocation (model 2) �Objective: max min �Subject to �Flow <= Capacity �Per OD pair average delay Dw ≦ D 45 2020/11/23

Joint flow control and routing (Model 1) 46 2020/11/23

Joint flow control and routing (Model 1) 46 2020/11/23

Joint flow control and routing (Model 2) 47 2020/11/23

Joint flow control and routing (Model 2) 47 2020/11/23

Joint flow control and routing (Model 3) J: average packet number upper bound 48

Joint flow control and routing (Model 3) J: average packet number upper bound 48 2020/11/23

Capacity allocation Flow control Routing Fairness criteria Y 49 N 2020/11/23

Capacity allocation Flow control Routing Fairness criteria Y 49 N 2020/11/23

discussion �model 2 v. s. model 3 �Multi-path �比較其他人的WORK (誰人較優) (pareto optimal) 50 2020/11/23

discussion �model 2 v. s. model 3 �Multi-path �比較其他人的WORK (誰人較優) (pareto optimal) 50 2020/11/23

Problem Complexity �Hassien function �Non-convex != NP hard 51 2020/11/23

Problem Complexity �Hassien function �Non-convex != NP hard 51 2020/11/23

Motivation �為何要討論 WMNs 的 End-to-End delay fairness? �引用文獻[][][] -他們說這是一個值得作的ISSUE �在同一gateway服務的區域,不該因距離遠近而效 能有所差異。 �We guarantee min

Motivation �為何要討論 WMNs 的 End-to-End delay fairness? �引用文獻[][][] -他們說這是一個值得作的ISSUE �在同一gateway服務的區域,不該因距離遠近而效 能有所差異。 �We guarantee min fair performance and try to reach as good as we can. 52 2020/11/23

Literature review (background) �Wireless mesh networks: a survey �End-to-end performance and fairness in multihop

Literature review (background) �Wireless mesh networks: a survey �End-to-end performance and fairness in multihop wireless backhaul networks �Knight 怎麼作fairness - Reference model 53 2020/11/23

Literature review (related) �Location-Dependent Throughput and Delay in Wireless Mesh Networks (廖老師WORK) �在routing已固定的情況下,以queueing system分

Literature review (related) �Location-Dependent Throughput and Delay in Wireless Mesh Networks (廖老師WORK) �在routing已固定的情況下,以queueing system分 析距離sink遠或近該如何妥善的配置頻寬已達到 fairness。 54 2020/11/23

Literature review (Lin) �Delay Constrained Routing and Link Capacity Assignment in Virtual Circuit Networks

Literature review (Lin) �Delay Constrained Routing and Link Capacity Assignment in Virtual Circuit Networks (2004 IEICE) �Three algorithm for routing and flow control in virtual circuit networks �Near-optimal Delay Constrained Routing in virtual circuit networks �Near-optimal tree-based access network design 55 2020/11/23

Literature review (Lin) �演福學長博論 56 2020/11/23

Literature review (Lin) �演福學長博論 56 2020/11/23