Yigal Bejerano and Randeep S. Bhatia Bell Laboratories, Lucent Technologies IEEE INFOCOM, 2004

1. MiFi：A Framework for Fairness and Qos Assurance in Current IEEE 802.11 Networks with Multiple Access Points Yigal Bejerano and Randeep S. Bhatia Bell Laboratories, Lucent Technologies IEEE INFOCOM, 2004

2. Outline • Introduction • System Goals • Overview of the MiFi System • The Beacon Block • The Slot Assignment Mechanism • The Admission Control • The Frequencies and the Slot Assignment Algorithm • Simulation • Intra-AP fairness • Inter-AP fairness • Overall system throughput • Conclusion

3. Outline • Introduction • System Goals • Overview of the MiFi System • The Beacon Block • The Slot Assignment Mechanism • The Admission Control • The Frequencies and the Slot Assignment Algorithm • Simulation • Intra-AP fairness • Inter-AP fairness • Overall system throughput • Conclusion

4. Introduction • MiFi, managed WiFi, is a framework for providing fair service and supporting Qos requirements in IEEE 802.11 networks with multiple access-points(APs). • Motivation：802.11 Limitations • DCF：support only best-effort services and cannot provide any guarantee on message delays. • PCF：provides certain degree of fairness in the case of a single AP, but cannot guarantee in networks with several APs.

5. Fairness and Qos • Fairness：the ability of a network to provide the same level of service to all its users • Qos：the ability of providing a service with some level of assurance for data delivery • Assurance：guaranteed bandwidth, delay bounds and jitter • Two related problems：a system cannot provide a certain degree of fair service to its users, i.e., minimal allocated bandwidth, cannot provide Qos guarantees.

6. Collisions Due to Two Problems • Hidden node problem • A station is called hidden when it is in the sensing range of the intended receiver but out of the sensing range of the transmitter. • Thus, a transmission of the hidden station may prevent the receiver from decoding the intended message. • Overlapping cell problem • Interference during transmissions in a CFP due to the transmissions in adjacent cells.

7. Hidden node problem Severe hidden nodes effect near boundaries

8. Outline • Introduction • System Goals • Overview of the MiFi System • The Beacon Block • The Slot Assignment Mechanism • The Admission Control • The Frequencies and the Slot Assignment Algorithm • Simulation • Intra-AP fairness • Inter-AP fairness • Overall system throughput • Conclusion

9. System Goals • To provide fair service to all its mobile users and to ensure Qos guarantees to real-time sessions, while maximizing the achievable overall network throughput

10. Goal of Fairness • Ensure that users of a single type (RT, NRT) experience the same network usage and the network resources are proportionally allocated among the users of the two types. • Spatial Fairness：the experienced service level should be independent of the distance between the users and their associated APs

11. Inter-AP and Intra-AP Fairness • Intra-AP Fairness： • balance between resource allocation to its NRT and RT-users according to a given fairness criteria. • Inter-AP Fairness： • when the efficient bandwidth Bv of an AP v is directly proportional to the total number of users mv associated to it. • i.e the ratio Bv/mv is the same for all APs

12. Example 1 and 2 • Achieving intra-AP or inter-AP fairness is contradictory to achieving high throughput Intra-AP fairness Inter-AP fairness Bv=Bu =1Mbps

13. Example 1：Intra-AP Fairness • Consider two APs u,v, user w is associated with AP v. Due to the large distance between the two APs, both of them can simultaneously exchange messages with their adjacent users except user w. • When v is exchanging messages with user w then u and all its associated users must be silent. • The network throughput is maximized by starving user w. • To maximize the network throughput, the intra-AP fairness(spatial fairness) requirement must be violated.

14. Example 2： Inter-AP Fairness • Two APs u,v do not interfere with each others communication and both have the same efficient bandwidth.(assuming Bv=Bu =1Mbps) • Inter-AP fairness requires that all the users in the system experience the same flow allocation, 1/5, since AP u has 5 associated users. • However, the network utilization can be increased by allowing increased flow allocation of ½, to the users associated with AP v without affecting the flow allocations of the users attached to AP u.

15. Outline • Introduction • System Goals • Overview of the MiFi System • The Beacon Block • The Slot Assignment Mechanism • The Admission Control • The Frequencies and the Slot Assignment Algorithm • Simulation • Intra-AP fairness • Inter-AP fairness • Overall system throughput • Conclusion

16. MiFi

17. Overview of the MiFi System • PCF mode can efficiently support RT sessions and provide fairness in WLAN within a single AP. • Extend PCF to multiple APs • Partition time into repeated periods or superframes • Each superframe has a fix length D and it contains a CFP followed by a CP

18. MiFi：Superframe=CFP+CP

19. NOC • Network operation center(NOC)：determines the slot assignment and synchronizes the APs • Each AP manages its own admission control for accepting new RT-sessions and determines its polling list. • A special software is used to control AP’s behavior for providing Qos and fairness to the attached users and for communicating with the NOC. • No modification of the IEEE 802.11 standard

20. CFP and CP • CFP ：for data transmission of both RT and NRT sessions • CP：a signaling channel for initiating new sessions and sending management messages. • The proportion of time allocated to each period is determined by the system needs to balance between fairness and network throughput.

21. MiFi： BB and EB

22. BB and EB • CFP starts with a beacon block(BB) in which all the APs transmit ‘almost’ at the same time beacon messages for initiating a CFP. • It ends with an end block(EB) in which all the APs send CF-end messages approximately at the same time to end their CFPs.

23. MiFi：Beacon Block = Jamming + Beacon Transmission Phase beacon transmission phase

24. The Beacon Blocks：Two Phases • The BB contains two phases：a jamming phase followed by a beacon transmission phase. • The jamming phase silences the network for a period of EIFS. • In the beacon transmission phase, APs send their beacon messages which will not suffer from collisions with messages transmitted by mobile user operating in DCF mode.

25. The Beacon Transmission Phase • Beacon messages from two interfering APs may collide, the beacon transmissions of APs are synchronized such that two adjacent APs in the interference graph do not send their beacon messages simultaneously. • For reducing the overhead of the beacon block, we would like to send the beacon messages as quickly as possible.

26. AP AP station station RT RS RT i u w j Interference Graph • Definition：G(V,E), is defined by the set V of APs and a set of edges E between every pair of APs u,v є V that are at most 2RT+RS apart, i.e., d(u,v) <= 2RT+RS • Transmission range(RT)： • the zone in which any message sent by station v can be correctly decoded. • Sensing range (RS)： • any station included in the range can sense every transmission. [18] T.S. Rappaport. Wireless Communication Principle and Practice. Rrentice Hall, 1996.

27. Graph Coloring Problem • Thus, we map the beacon synchronization problem into a graph coloring problem that seeks to find the minimal number of colors that are needed to color the interference graph, such that all the nodes with the same color send their beacon messages simultaneously.

28. Example 3 • Since G=(V,E) is 3-colorable, the beacon block contains 3 beacon slots. • First：a,d；Second：b,c,f；Third：e

29. MiFi：Beacon Transmission Phase Slot Assignment beacon transmission phase

30. MiFi：CFP Slot Assignment

31. The Slot Assignment Mechanism • Goal：Maximize the network throughput while ensuring inter-AP fairness. • CFP：can be divided into R slots enumerated from 1 to R. • Sv：the set of slots that are allocated to AP v • rv：the number of slots in Sv.

32. The Slot Assignment Mechanism • A slot assignment is a vector S={Sv1,Sv2 …Sv|V|}, of the sets Svi for every AP viєV. • A slot assignment is termed feasible if for every AP v, Sv [1…R] and any pair of adjacent nodes in the interference graph G(V,E) do not have any common slot. i.e., for every (u,v) є E, it follows that Su∩ Sv = ø • A feasible slot assignment S is optimal if it maximizes the min-slot-to-user ratio defined by ρ= minv єV (rv / mv) ∩

33. The Efficient Approximation Algorithm • The first is the coloring algorithm that given a graph G(V,E) and the number of colors, rv ,required by every node v є V, finds a feasible color assignment with minimal number of colors. (later describe) • It performs a binary search for finding the maximal min-slot-to-user ratio ρ that requires no more than R slots.

34. The Efficient Approximation Algorithm • At each iteration, it selects a ratio ρ and sets the requirement of every node v єV to rv = ρ. mv colors. • The algorithm then uses the coloring algorithm to check whether there is a feasible slot assignment with R slots(colors). • The algorithm picks lower or higher value for the ratio ρuntil it quickly converges to the optimal ratio ρ.

35. Example 4：Slot Assignment Given user number per AP, and each superframe contains 5 slots • ρ=1 • r1 = 1· 9 > 5 • ρ=1/2 • r1 = 1/2· 9 = 5 • ρ=1/4 • r1 = 1/4· 9 =3 • r3 = 1/4· 10 =3 • ….. • ρ=1/5 optimal

36. Example 4 • A slot assignment that maximizes the min-slot-to-user ratio of the given interfering graph G(V,E). Each superframe contains 5 slots and the number of users mv attached to each AP v is depicted near each node v. • The figure shows the allocated slots Sv as well as the slot-to-user ratio |Sv|/Mv of each node v. In this case the maximal min-slots-to-user ratio is 1/5 due to node c and e.

37. Admission Control • For ensuring intra-AP fairness, each AP employs an admission control mechanism. • Consider an AP v that has rv slots and is associated mv users, where mvRT of them are RT users. Let ∆ be the number of time units in every slots. • Balances between success probability of RT-sessions requests versus the average flow given to each NRT-users.

38. Admission Control – RT users • New RT-session request are approved only while the aggregated RT flow does not exceed a threshold of Hv = c‧(mvRT / mv)‧rv‧∆ , for a given configuration parameter c ≧1 and a requirement that Hv< rv‧ ∆. • An RT-user initiates a new RT-session by sending a request to its AP during CP. If the AP approves the request then it allocates a time unit to this user and adds the users address to its polling list.

39. Admission Control – NRT users • In the CFP the AP first polls all the RT-users with active RT-sessions and in the remaining time of its slots it polls its NRT-users. • For ensuring intra-AP fairness, the AP employs a sliding window method for determining the next NRT-user to poll at time t. The AP keeps records of the number of successfully served messages by each NRT-user until now. • The polled NRT-user is the one which has the minimal number of served messages during that time period.

40. Polling list

41. Outline • Introduction • System Goals • Overview of the MiFi System • The Beacon Block • The Slot Assignment Mechanism • The Admission Control • The Frequencies and the Slot Assignment Algorithm • Simulation • Intra-AP fairness • Inter-AP fairness • Overall system throughput • Conclusion

42. The Frequencies and Slot Assignment Algorithm • n(v)：the set of neighbors of node v in the graph G • v1,v2,…vn：the nodes of G ordered in non-decreasing X-coordinate of their locations • Algorithm processes the vertices in the reverse order vn,vn-1…v1 and uses a generalized First-Fit for its frequency and color assignment.

43. The Frequencies and Slot Assignment Algorithm • N(vi)n(vi)：the set of neighbors of node vi among the nodes vi+1, vi+2…vn. Thus N(vn) = ø. • When vi is considered by A, all the nodes in N(vi) have already been assigned colors and frequencies by the algorithm. • Nf(vi)N(vi) be the set of neighbors of node vi in N(vi) that have been assigned frequency f є F by A. |∩ |∩

44. The Frequencies and Slot Assignment Algorithm • Assuming frequency f, the algorithm computes the least colors rvi that can be assigned to v, while considering only the nodes in Nf(vi). • Then, the algorithm A selects the frequency f for which the maximum color assigned to vi by First-Fit is minimized and assigns, accordingly, the set of color to node vi.

45. The Frequencies and Slot Assignment Algorithm

46. Example 5：Node f, e F consists of 2 frequencies：f1 and f2

47. Example 5

48. Example 5

49. Example 5 • Steps： • Assigns color 1 and frequency f1to node f • For node e, the smallest available color for frequency f1is 2 and for frequency f2is 1. Hence, algorithm assigns color 1 and frequency f2is 1. • The set of available colors for node d are all colors except 1 for both frequencies f1 and f2. Thus, A assigns colors 2 and 3 and frequencies f2 to node d. • Algorithm A assigns colors to node c and the minimum available colors 2 and 3 for f2 and 1 and 4 for frequency f1. Hence, A assigns colors 2 and 3 and frequency f2 to node c.

50. Example 5 • Steps： • For node b, the best frequency is f2 and for this frequency algorithm A assigns it color 1. • For node a the minimum available colors are 1 and 2 for frequency f1 and 1 and 4 for frequency f2. Hence, algorithm A assigns colors 1 and 2 and frequency f1 to node a. • Thus, the total number of colors used by algorithm A for G with two frequencies is 3,which is optimal.