Wireless Scheduling and Channel Assignment

1
Wireless Scheduling and Channel Assignment

• Presenter Gaurang Sardesai Xi Liu

2
Exploiting Medium Access Diversity (MAD) in Rate
Adaptive Wireless LANs

• Goal
• Exploit variations in channel conditions and
  improve overall network throughput
• How?
• Obtain instantaneous channel information from
  multiple receivers and selectively transmit data
  to receiver that improves overall throughput of
  network
• Maintain temporal fairness.

3
Rate Adaptivity and Multi User Diversity

• Rate bounded by signal to noise ratio which
  changes over time.
• Upper layers should react quickly to changes in
  channel conditions
• What is multi user diversity?
• Instantaneous channel conditions are time varying
  and not correlated
• Aggressively assess channel conditions and select
  receiver whose channel conditions are at the peak
• If experiencing failures, switch users
• Not only best user, but best rate as well
• So why cant you apply this algorithm to DCF?

4
Overview
Sender
Receivers
Data
Query
Reply
5
What do you need to take care off?

• Overhead of Probing
• Maximize throughput, especially when conditions
  are favorable
• Be fair between multiple traffic flows. This
  usually conflicts with previous objective.
• 3 phases
• Channel Probing
• Data Transmission
• Receiver Scheduling

6
Channel Probing

• Group RTS
• CTS
• 2 additional Fields
• Rate
• Gain
• Probing concludes
• Problems?
• Duration Field
• Conservative Estimate

7
Data Transmission

• Might as well send as many as you can
• OAR
• Low data rate for hidden terminal problem
• SIFS fixed duration
• Enter PAC
• Transmit SuperFrame, followed by string of Data
  frames
• Receiver waits for SIFS, sends group ACK
• Number of Packets should not exceed ratio of
  current to base rate for fairness
• Retransmission counter for each packet
• SF contains RA bitmap. ACK also modified

8
Data Transmission contd
9
Receiver Scheduling

• Choose node with maximum relative gain for each
  transmission phase.
• Maximum relative gain scheduling has temporal
  fairness, and the difference in throughput is
  bounded.
• As number of receivers increase, overhead
  increases. So two approximations proposed for
  ideal scheduling algorithm.
• K-set round robin
• Revenue Based

10
Performance Analysis

• How many do you want to query?
• Optimal value 3
• Network Throughput
• Compare OAR, PAC and DCF

11
Performance Evaluation
12
Performance Evaluation w.r.t. Topology
13
Fairness and Load Balancing in Wireless LANs
Using Association Control

• Motivation
• User associates with AP that has strongest RSSI,
  ignoring the load
• Load is unevenly distributed among APs
• Unfair bandwidth allocation among users
• Goal
• Balanced load and fair bandwidth allocation
• Basic idea
• Association control (user-AP association) to
  ensure max-min fairness bandwidth allocation and
  min-max load balancing

14
Basic idea

• Each user monitors the signal strength of beacons
  from nearby APs
• Measures the effective bit rate
• Clients submit this information to a network
  control center (NOC)
• NOC runs scheduling and decides users
  associations
• Users switch association accordingly

15
Single association vs. Fractional association
Infrastructure
Infrastructure
AP1
AP2
AP2
AP1
Single association
Fractional association
16
Max-min fairness

• Informally
• if there is no way to give more bandwidth to any
  user without decreasing the allocation of another
  user with less or equal bandwidth
• Formally
• Allocation vector Bb1,,bn, bi is the
  bandwidth allocated to user i
• Lexicographically largest feasible allocation

17
Example of max-min fairness
2
2
2
1
4
2
2
2
1
B 1, 1, 1, 1, 1
b1 for each user
Wireless System
2
1
2
1
4
4
2
2
4
4
2
B 1, 4/3, 4/3, 4/3, 4/3
B 1, 1, 1, 2, 2
Max-min fairness fractional association
Max-min fairness single association
18
Load

• What is a good indicator of load?
• Number of users associated? (X)
• Throughput of AP? (X)
• Intuitively
• the load of an AP needs to reflect its inability
  to satisfy the requirements of its associated
  users
• it should be inversely proportional to the
  average bandwidth that it experiences

19
Load (contd.)

• Each client associates with an AP fractionally
• E.g. Node n1 associates with AP1 1/2 of the time,
  and effective data rate is 3Mbps
• The load a client poses on an AP
• E.g. Node n1 induce a load of 1/6 s/Mb on AP1
• The load on AP is the sum of loads from
  associated clients

20
Example of min-max load balance
2
2
2
Y 1, 1, 1
2
2
1
4
2
1
B 1, 1, 1, 1, 1
b1 for each user
Wireless System
2
Y 1, 3/4, 3/4
1
2
Y 1, 1, 1/2
1
4
4
2
2
4
4
2
B 1, 4/3, 4/3, 4/3, 4/3
B 1, 1, 1, 2, 2
Max-min fairness fractional association
Max-min fairness single association
21
Relationship of max-min fairness and min-max load
balance

• In the fractional association case, a min-max
  load balanced association X defines a max-min
  fair bandwidth allocation and vice versa.
• However, the theorem is not satisfied in the case
  of a single association.

Infrastructure
Infrastructure
APa
APb
APc
APc
APa
APb
2
1
4
4
2
2
1
4
4
2
Y 1,1,1/2
Y 1,1,1/2
B 1,1,1,2,2
B 1,1,1,1,2
22
Integral load balancing

• It is NP-hard
• Step 1 Finding optimal fractional association
• In each iteration, identify bottleneck access
  points and users
• Remove them and start the next iteration
• This algorithm yields a min-max load balanced
  association
• Step 2 Rounding to obtain approximate integral
  association

23
Bottleneck detection

• Calculates an fractional association that
  minimizes the maximum load on all APs
• Use linear program to minimize bottleneck load
• It only optimizes bottleneck, but not other APs
• Minimize sum of load on all APs, given bottleneck
  load (Identify those APs in bottleneck load
  group)
• Use another linear program
• Build a directed graph to see whether load can be
  shifted from one AP to another

24
Example of bottleneck detection
b
c
a
a
b
c
A possible association calculated by LP2
25
Simulation

• Compare with Strongest-Signal-First and
  Least-Loaded-First
• User effective bit rate only depends on distance
  only
• Backhaul capacity is 10Mbps
• Transmission range is 150m
• 20 APs
• 5 4 grid
• Inter-AP distance is 100m
• 100 users

26
Results Per-user bandwidth
27
Summary

• Consider fairness in conjunction with load
  balancing
• "In the presence of hotspots, our algorithms
  provide fair service to all users accessing the
  network, while also maximizing the amount of
  bandwidth they receive," said Yigal Bejerano, a
  researcher in Bell Labs' Internet Management Lab.
  Bejarano continued, "Typically our algorithms
  also yield higher network utilization than the
  most commonly used 'strongest signal approach,
  while today's approaches tend to focus on overall
  throughput when allocating network resources. We
  believe that understanding the correlation
  between fairness and load-balancing are critical
  in order to maximize bandwidth for all users."

28
Coordinated Load Balancing, Handoff/Cell-site
Selection, and Scheduling in Multi-cell Packet
Data Systems

• Motivation
• Inter-cell interference
• Asymmetric load distribution
• Goal
• Improve global resource utilization
• Reduce regional congestion
• Basic idea
• Packet-level scheduling
• Call-level cell-site selection and handoff
• System-level load balancing

29
Model

• Entities
• Central server
• Base station (BS)
• Mobile station (MS) minRate requirement
• Link model
• Path loss
• Fast Reyleigh fading
• Slow shadowing fading
• Channel rate depends on SINR

30
System Coordination

• Mobile Station
• Channel strength at from each BS
• Number of active users at each BS
• Choose the optimal serving BS
• Constantly measure average throughput for
  load-aware handoff
• Base Station
• Broadcasts mean number of its binding MSs
• Periodically updates load to a central controller
• Central Controller
• Executes centralized tuning of cell coverage
  (Cell breathing)

31
Example
32
Packet-level scheduling

• Assignment problem
• Goal is to maximize the long-term revenue
• At each timeslot, each BS can choose any MS to
  serve
• MS can be served by at most one BS at a time
• Problems
• Require fine-grained global knowledge
• The computation is required for each timeslot
• Suboptimal solution
• Each MS binds to a BS (dynamic binding)
• Each cell schedule by BS

33
Cell-site selection (MS)

• Cross-layer scheme
• Instead of merely SINR-based
• Goal is to maximize the net increment of utility
• New utility
• - Utility drop by other competing stations
• Estimate new throughput
• Rate / Num of users
• BS accepts admission of MS if and only if total
  capacity after accepting the MS does not exceed 1
• Conservative but robust

34
Weighted Alpha Rule (BS)

• Assignment problem inside a cell
• Utility function to achieve (w,a)-proportional
  fairness
• w is weight
• a is a tuning knob balancing fairness and
  aggregate throughput
• a 0, scheduler is biased toward maximum
  throughput
• a 1, scheduler assigns slots equally
• Minimum rate requirement
• Tune weight

35
Cell breathing (Controller)

• If a cell is more congested than its neighbor
  cells, it reduces a
• Reducing a makes the scheduler to bias toward
  fast station
• BS will allocated less slots to MSs at cell
  boundary
• Boundary MSs will monitor less throughput and may
  trigger handoff
• Effectively the cell coverage is reduced
• Load is defined to be the ratio of minimum
  required rate to actual data rate

36
3-tier cell system
37
Analysis - dynamics
38
Analysis - performance
39
Performance Anomaly of 802.11b

• Useful throughput is much smaller than nominal
  bit rate
• 7.74Mbps vs. 11Mbps
• Contention time strongly depends on number of
  contending hosts
• Fast host obtain the same throughput as slow host
• Slow host will may considerably limit throughput

40
Facilitating Access Point Selection in IEEE
802.11 Wireless Networks

• Basic idea
• The bandwidth an end-host is likely to receive if
  it were to affiliate with a given access point
• Use timing to estimate the load on AP and the
  contention inside the network

41
Experimental results
42
Questions

• What is the difference between load on access
  point in wireless network and load on load on
  routers in wired network?
• If the users always associate with the AP with
  the highest throughput, will it lead to max-min
  fairness?

43
Reference

• Improving protocol capacity with model-based
  frame scheduling in IEEE 802.11-operated WLANs,
  Proceedings of the 9th annual international
  conference on Mobile computing and networking,
  ACM, San Diego, CA, USA
• Yigal Bejerano Seung-Jae Han and Li (Erran) Li,
  Fairness and Load Balancing in Wireless LANs
  Using Association Control, Proc. International
  Conference on Mobile Computing and Networking
  (MobiCom), Philadelphia, PA, September 2004.
• Exploiting Medium Access Diversity in Rate
  Adaptive Wireless LANs Z. Ji, Y. Yang, J. Zhou,
  M. Takai and R. Bagrodia. To appear in
  Proceedings of ACM MOBICOM 2004, Philadelphia,
  Sep 26 - Oct 1, 2004.
• Martin Heusse, Franck Rousseau, Gilles
  Berger-Sabbatel, and Andrzej Duda. Performance
  Anomaly of 802.11b. In Proc. of IEEE INFOCOM,
  March 2003
• Victor Bahl, Ranveer Chandra, and John Dunagan.
  SSCH Slotted Seeded Channel Hopping for Capacity
  Improvement in IEEE 802.11 Ad-Hoc Wireless
  Networks. Proc. of ACM Mobicom 2004, Sept.-Oct.
  2004.
• Ashish Raniwala and Tzi-Chiueh. Architecture and
  Algorithms for an IEEE 802.11-based Multi-channel
  Wireless Mesh Network In Proc. of IEEE INFOCOM,
  March 2005.
• Coordinated Load Balancing, Handoff/Cell-site
  Selection, and Scheduling in Multi-cell Packet
  Data Systems, Aimin Sang, Xiaodong Wang, Mohammad
  Madihian, Richard Gitlin, ACM Mobicom 2004.
• Facilitating Access Point Selection in IEEE
  802.11 Wireless Networks, S. Vasudevan, K.
  Papagiannaki, C. Diot, J. Kurose, and D. Towsley,
  In ACM Internet Measurement Conference, 2005

44
Q A

• Thanks!