Title: Throughput Optimization and Fair Bandwidth Allocation in MultiHop Wireless Local Area Networks
1Throughput Optimization and Fair Bandwidth
Allocation in Multi-Hop Wireless Local Area
Networks
INFOCOM 2006
- Qunfeng Dong (University of Wisconsin-Madison)
- Suman Banerjee (University of Wisconsin-Madison)
- Benyuan Liu (University of Massachusetts-Lowell)
Slides by Ting-Yu Lin
2Outline
- Introduction
- The Performance anomaly of 802.11 WLANs
(one-hop) - Previously proposed one-hop/multi-hop solutions
and limitations - Our multi-hop solution
- Fairness in multi-hop WLANs
- Max-Min Throughput Fairness
- Max-Min Time Fairness
- Fair bandwidth allocation in multi-hop WLAN
- Construction and maintenance of an efficient
tree - Backward compatibility
- Evaluation
- Summary
3Background
- The Performance anomaly of 802.11 WLANs
- Channel rate diversity is prevalent in 802.11
WLANs KNE03, TG04 - 802.11 DCF implements Throughput-based Fairness
- Assigns approximately equal bandwidth to each
client - A low bit rate client brings down throughput of
everyone HRBD03
Rate of 8,9 11 Mbps Rate of 5,6,7 5.5
Mbps Rate of 1,2,3,4 2 Mbps Total
Throughput 3.3 Mbps
4Background
- Solution 1 Time-based Fairness TG04
- Each client receives equal time share
- High bit rate clients are protected
- Aggregate throughput is increased
- However, throughput of low bit rate clients is
decreased!
Rate of 8,9 11 Mbps Rate of 5,6,7 5.5
Mbps Rate of 1,2,3,4 2 Mbps Total
Throughput 5.2 Mbps
5Background
- Solution 2 Multi-AP Association Control BHL04
- In 802.11, clients are assigned to the AP with
the strongest RSSI - Often leads to unbalanced load distribution and
idle channel - Throughput and fairness can be improved by
careful association
6Background
- Solution 2 Multi-AP Association Control BHL04
- In 802.11, clients are assigned to the AP with
the strongest RSSI - Often leads to unbalanced load distribution and
idle channel - Throughput and fairness can be improved by
careful association
Bandwidth of client 1 1 Mbps Bandwidth of
client 2 1 Mbps Bandwidth of client 3 1
Mbps Bandwidth of client 4 1 Mbps Bandwidth of
client 5 1 Mbps
A fair default association
7Background
- Solution 2 Multi-AP Association Control BHL04
- In 802.11, clients are assigned to the AP with
the strongest RSSI - Often leads to unbalanced load distribution and
idle channel - Throughput and fairness can be improved by
careful association
Bandwidth of client 1 1 Mbps Bandwidth of
client 2 2 Mbps Bandwidth of client 3 2
Mbps Bandwidth of client 4 1 Mbps Bandwidth of
client 5 1 Mbps
A max-min fair association
8Motivation Challenges
- A common limitation of these previous techniques
- Only use AP-client links, which often have low
bit rates. - High rate P2P links between clients exist, but
not utilized. - Solution 3 Multi-Hop WLAN LBB04
- Forward traffic via high bit rate P2P links
between clients - Less channel access time is needed ? increased
throughput - Challenges
- How to build an efficient multi-hop tree rooted
at the AP? - How to conduct fair bandwidth allocation in a
multi-hop topology? - Objective
- Improve throughput without sacrificing fairness
in multi-hop WLAN
9Our Results
- Max-min throughput fair allocation in multi-hop
WLANs - No client can get more bandwidth without
decreasing the bandwidth of clients with the same
or less bandwidth - Max-min time fair allocation in multi-hop WLANs
- No client can get more time share without
decreasing the time share of clients with the
same or less time share - Multi-hop tree construction
- Iteratively search for a better tree topology
such that fair bandwidth within the new tree
leads to better client throughput
10Fairness
- Max-min throughput fairness
- A feasible bandwidth allocation B is max-min
throughput fair if - No client can be allocated more bandwidth without
decreasing the bandwidth of clients with the same
or less bandwidth - Max-min time fairness
- A feasible bandwidth allocation B is max-min time
fair if - For each node i in the multi-hop tree, assume the
same amount of total bandwidth allocated to nodes
in the sub-tree rooted at node i - Neither node i nor any child of it can be
assigned more time share of node i without
decreasing the time share of some of them with
equal or less time share of node i - Gives more bandwidth to forwarding nodes and
hence - Encourages clients to serve as forwarding nodes
- Leads to more skewed bandwidth allocation but
higher throughput
11Methodology
Focus on the up-link direction, and solve the
bandwidth allocation within a single-AP tree first
More aggregate throughput 55/6
Aggregate throughput 44/5
12Max-min throughput fairness (MMFA)
A bandwidth allocation B is feasible if and only
if it is feasible at every node i with
Pi
i
Qij1, j2, j3
j3
j2
j1
13Max-min throughput fairness (MMFA)
General idea
- Fair bandwidth allocation within a multi-hop tree
structure is recursively done in a bottom-up
order - During the bottom-up recursive process, at each
node i, MMFA performs the pump-and-drain
operation - First pump the node, i.e., allocate to the node
a bandwidth such that it is fully loaded or even
overloaded - Then drain the nodes in the sub-tree rooted at
that node, i.e., decrease their allocated
bandwidth such that the resulting bandwidth
allocation is feasible and throughput fair within
the sub-tree -
14Max-min throughput fairness (MMFA)
Detailed design
- Each node i maintains and reports to its parent
the following info. -
15Max-min throughput fairness (MMFA)
16Max-min throughput fairness (MMFA)
17- Original case of max-min throughput fair
allocation - in a single-hop WLAN
-
r8, r9 11 Mbps r5, r6, r7 5.5 Mbps r1, r2, r3,
r4 2 Mbps
- 2bi/11 3bi/5.5 4bi/2 1
- 4bi 12bi 44bi 22
- bi 11/30
- aggregate throughput 99/30 3.3 Mbps
18- Example of max-min throughput fair allocation
using MMFA - in a multi-hop WLAN (all links have a 11 Mbps
data rate) -
At node 7 Pump b711 W711/112(11/11)3 gt
1 Drain n4T4nN7L71 d(3-1)/(1/11n4/11n4/1
1)22/3 b7b411-22/311/3
At node 5 Pump b511 W511/114(11/11)5 gt
1 Drain n1T1nN5L51n2 d(5-1)/(1/112(n1/1
1n1/11))44/5 b5b1b211-44/511/5
19At node 9 Pump b911/3, B722/3
W91/3B7/11B7/115/3 gt 1 Drain
n7T7nN9L92 d(5/3-1)/(1/11n7/11n7/11)22/
15 b9b7b411/3-22/1511/5
At node 8 Pump b811/3, B533/5, B622/3
W81/323/522/343/15 gt 1 Drain
n6T6nN8L82 d(43/15-1)/(1/11n6/11n6/11)
4211/75 gt L8L8- L8L8-1 gt
b8b6b3b5b1b211/5 Drain again
W81/52(3/52/5)11/5 gt 1 d(11/5-1)/(1/115/115
/11)6/5, so 11/5-6/51
20Finally Drain at the AP B86, B933/5 WAP6/113/
563/55 gt 1, n93 d(63/55-1)/(n9/11)8/15 Thus
b9b7b411/5-8/155/3
Aggregate throughput is maximized 11Mbps
21Max-min time fairness
t
Pi
i
j
Tj 3
22Max-min time-based fairness allocation (TBFA)
General idea
- Fair bandwidth allocation within a multi-hop tree
structure is recursively done in a bottom-up
order - During the bottom-up recursive process, at each
node i, TBFA performs the pump-and-drain
operation - First pump the node, i.e., allocate to the node
a bandwidth such that it is fully loaded or even
overloaded - Then drain the nodes in the sub-tree rooted at
that node, i.e., decrease their allocated
bandwidth such that the resulting bandwidth
allocation is feasible and time fair within the
sub-tree -
23Max-min time-based fairness allocation (TBFA)
24Max-min time-based fairness allocation (TBFA)
25- Original case of max-min time fair allocation
- in a single-hop WLAN
-
r8, r9 11 Mbps r5, r6, r7 5.5 Mbps r1, r2, r3,
r4 2 Mbps
- time share at each node 1/9
- b8b911(1/9)11/9
- b5b6b75.5(1/9)11/18
- b1b2b3b42(1/9)2/9
- aggregate throughput 5.2 Mbps
26- Example of max-min time fair allocation using
TBFA - in a multi-hop WLAN (all links have a 11 Mbps
data rate) -
At node 7 subtree size G72 Time share
T77T741/2 Bp4T74/(1/111/11)11/4b4
b7111/211/2
At node 5 subtree size G53 Time share
T55T51T521/3 Bp1T51/(1/111/11)11/6b1 Bp2Bp
111/6b2 b5111/311/3
27 Drain node 7 subtree size G41 d(33/4-11/3)/(1
1G4/(1/111/11)) 5/18 Time share T77
T741/2-5/182/9 b7112/922/9,
b4(11/2)(2/9)11/9
At node 9 Time share T991/3, T972/3 Bp7T97/(1/
111/11)11/3 b9111/311/3 B711/42/1133/4
gt Bp7
28At node 8 Time share T881/6, T861/3,
T851/2 Bp6T86/(1/111/11)11/6 lt
B633/4 Bp5T85/(1/111/11)11/4 lt B522/3
b8111/611/6
Drain node 5 G1G21 Time share
T55T51T521/3 d(22/3-11/4)/(112/(1/111/11))
5/24 Time share T55T51T521/3-5/241/8
b5111/811/8 b1b2(11/2)(1/8)11/16
Drain node 6 G31 Time share T661/2,
T631/2 d(33/4-11/6)/(11G3/(1/111/11)) 7/18
Time share T66 T631/2-7/181/9
b6111/911/9, b3(11/2)(1/9)11/18
29 Drain node 9 Time share T991/3, T972/3 with
G72 d(22/3-55/12)/(11G7/(1/111/11))1/8 Ti
me share T991/3-1/85/24 b911(5/24)55/24 Tim
e share T972/3-2(1/8)5/12 Bp7T97/(1/111/11)5
5/24 lt B711/3 Drain node 7 (continued in next
slide..)
Finally Drain at the AP Time share TAP91/3,
TAP82/3 Bp9TAP9/(1/11)11/3 lt
B922/3 Bp8TAP8/(1/11)22/3 gt B877/12 TAP8(77/1
2)/117/12 No Drain needed at node
8! TAP91-7/125/12 Bp911(5/12)55/12 lt B922/3
30 Drain node 7 Time share T77T742/9 with
G41 d(11/3-55/24)/(11G4/(1/111/11))1/12 T
ime share T77T742/9-1/125/36
b7115/3655/36 b4(11/2)(5/36)55/72
Aggregate throughput is maximized 11Mbps
31Tree Construction
- Start with a default WLAN topology
- For example, a one-hop association based on RSSI
- Iteratively search for a better tree topology
- Such that fair bandwidth allocation within the
new tree leads to a better client bandwidth
vector - Until no better tree can be found
- This iterative search can also be used to handle
client arrival and client leave - If a client arrives, quickly associate it with
some AP - For example, the AP with the strongest RSSI
- If a client leaves, quickly associate its
children with some AP - Then iteratively search for a better tree
topology
32- A good multi-hop tree structure
-
By migrating node 3 from node 6 to node 7!
Original tree structure
Equal bandwidth share at each node 11/9 And the
aggregate throughput is also maximized 11Mbps
33Backward Compatibility
- Legacy client devices can work as usual
- Upon arrival, simply associate with the AP with
the strongest RSSI - Not necessary to participate in tree construction
- Allocated bandwidth changes as the tree topology
evolves - Upon leave, simply disconnect from the AP
- No children, no additional processing needed
- Incremental deployment is enabled and encouraged
- Investing to upgrade client device will lead to
perceivable boost in performance, in spite of the
existence of other legacy client devices
34Evaluation
- Network settings
- 30 clients
- Scenario I 150m150m One AP at each corner
- Scenario II 300m300m One AP at each corner
- Scenario III 150m150m One AP at the center
- Scenario IV 300m300m One AP at the center
- Link rates
- 0m-50m 11 Mbps
- 50m-80m 5.5 Mbps
- 80m-120m 2 Mbps
- 120m-150m 1 Mbps
- 150m N/A
35Evaluation
36Evaluation
37Evaluation
38Evaluation
39Evaluation
40(No Transcript)
41Summary
- Unpractical interference model assumed (Truth
even though nodes are not associated with each
other, they still interfere!). - The extension part to the case of multiple APs is
weak (Trees need to be node-disjoint). - This work fails to address how to construct a
good multi-hop tree structure (only simple
iterative heuristic is suggested), which should
play a critical component in throughput
optimization and fairness provisioning.