the difference in tcp and udp results is due to the flow control mechanism of tcp

1
The difference in TCP and UDP results is due to the flow control mechanism of TCP. I T1 T2 0 1 2 Future Work •Implement different priority assignment strategies •Identify potential objectives to guide priority assignment •Ensure throughput regardless of route length by categorizing according to hops •Ensure throughput of certain users by categorizing according to source •Ensure throughput of certain applications by categorizing accord to packet type. •Devise a performance criteria to evaluate fairness Simulation Setup •Types of traffic •Constant Bit Rate traffic over UDP. •UDP is unreliable, one way traffic •FTP traffic over TCP •TCP is reliable two way traffic with flow control •Performance Metrics •Calculate end-to-end throughput for TCP •Calculate end-to-end success rate for UDP •Assumed error-free transmission •Link rate: 1Mbps •Five trials each Simulations Threshold is set equal to the probability of servicing others’ packets before your own. Motivation •Wireless networks of many different topologies are in use today for anytime, anywhere access •Multiple user accesses contend for network resources •Contention is primarily arbitrated in medium access control (MAC) and network scheduling •At each individual node, the network layer allocates the share of transmission time given by the MAC layer to different flows •To control QoS of contending flows, the project exploits priority queuing methods for network scheduling Typical Queuing Methods First-In-First-Out (FIFO) •Packets serviced depending on arrival time. Strict Priority •Packets placed into different queues according to some criteria (type, source…) •Service queue A, unless it’s empty. Then service queue B, and so on. Weighted Fair •Packets placed into different queues according to some criteria (type, source…) Service queue i for x i fraction of the time where and T = total number of queues. T i i x 0 1 Thomas Shen, University of Illinois-UC and Dr. K.C. Wang, Clemson University Quad Chain Pitfalls TCP throughput for multi-hop traffic on the quad chain and small mesh were terrible Lack of MAC access prevents packets from being sent With few packets, queuing method has no effect IEEE 802.11 MAC protocol is not efficient for multi- hop networks as documented in literature Thresholds T1 and T2 were both varied. T2 controls Flow 1 and 2. At T2 = 0.5 , UDP flows 1 and 2 got the same end-to-end rate. Conclusion •Results show throughput is unbalanced using FIFO •Priority queuing allocates bandwidth among flows •In our simulations, thresholds of 0.5 Figure 2 Figure 4 Our Priority Strategy Queue 0 Queue 1 Queue 2 Routing Packets Own Packets Others’ Packets Queue 0 Queue 1 Queue 2 If packets exist MAC layer If packets exist If packets exist else Probabilit y p Probability 1-p •Use a combination of strict priority and weighted fair queuing •Categorize packets based on packet type and source. •Service routing packets first, since routes needs to be established before other packets can reach their intended destination For example, the packet assignment for the triple chain UDP scenario is shown in Figure4. 3 3 2 1 2 0 ROUTING ROUTING ROUTING Q1 Q2 Q1 Q2 Q1 Q2 Figure 3 Figure 8 I T 0 1 3 2 Flows Triple Chain Figure 5 Figure 6 Figure 7 I 2 2 4 5 0 1 3 Small Mesh Figure1 0 Figure 11 In a wireless mesh network, routers are connected wirelessly as shown in Figure 1. Traffic from end-users travel through different routes as shown above. The different lengths and contention along each route affects performance of the flows. Routes that service multiple flows are places where prioritization strategies can be utilized to adjust performance. Interne t Figure 1 Threshold change allocated bandwidth between Flow 2,4 and 5. Other flows were unaffected due to lack of serious contention. Figure 9 UDP flows are 200Kbps CBR traffic Threshold Threshold 1 Threshold 2 UDP flows are 200Kbps CBR traffic UDP flows are 100Kbps CBR traffic Threshold changed for all intermediate nodes. 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 Triple C hain U D P R esults Flow 0 (1 hop) Flow 1 (1 hop) Flow 2 (2 hop) Flow 3 (2 hop) Flow 0 O riginal Flow 1 O riginal Flow 2 O riginal Flow 3 O riginal U D P E nd-to-end Success R ate Threshold 0 50 100 150 200 250 300 350 0 0.2 0.4 0.6 0.8 1 Triple C hain TC P R esults Flow 0 (1 hop) Flow 1 (1 hop) Flow 2 (2 hop) Flow 3 (2 hop) Flow 0 O riginal Flow 1 O riginal Flow 2 O riginal Flow 3 O riginal TC P E nd-to-end Throughput(Kbps) Threshold 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 Sm allM esh U D P R esults Flow 0 (1 hop) Flow 1 (2 hop) Flow 2 (2 hop) Flow 3 (2 hop) Flow 4 (3 hop) Flow 5 (3 hop) Flow 2 O riginal Flow 4 O riginal Flow 5 O riginal U D P End-to-end Success R ate Threshold

Upload: alyn

Post on 19-Jan-2016

37 views

Category:

Documents


2 download

DESCRIPTION

Priority Queuing : Achieving Flow 'Fairness' in Wireless Networks. I. One hop. 0. 3. 1. 2. 2. 1. 2. 0. Two hop. 5. 4. T2. T1. I. Routing Packets. Queue 0. Own Packets. Queue 1. Others’ Packets. Queue 2. Q1. Q1. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: The difference in TCP and UDP results is due to the flow control mechanism of TCP

The difference in TCP and UDP results is due to the flow control mechanism of TCP.

I T1 T2 0

1

2

Future Work•Implement different priority assignment strategies

•Identify potential objectives to guide priority assignment•Ensure throughput regardless of route length by categorizing according to hops•Ensure throughput of certain users by categorizing according to source•Ensure throughput of certain applications by categorizing accord to packet type.

•Devise a performance criteria to evaluate fairness

Simulation Setup•Types of traffic

•Constant Bit Rate traffic over UDP.•UDP is unreliable, one way traffic

•FTP traffic over TCP•TCP is reliable two way traffic with flow control

•Performance Metrics•Calculate end-to-end throughput for TCP•Calculate end-to-end success rate for UDP

•Assumed error-free transmission•Link rate: 1Mbps•Five trials each

Simulations

Threshold is set equal to the probability of servicing others’ packets before your own.

Motivation•Wireless networks of many different topologies are in use today for anytime, anywhere access•Multiple user accesses contend for network resources

•Contention is primarily arbitrated in medium access control (MAC) and network scheduling •At each individual node, the network layer allocates the share of transmission time given by the MAC layer to different flows

•To control QoS of contending flows, the project exploits priority queuing methods for network scheduling

Typical Queuing MethodsFirst-In-First-Out (FIFO)

•Packets serviced depending on arrival time.Strict Priority

•Packets placed into different queues according to some criteria (type, source…)•Service queue A, unless it’s empty. Then service queue B, and so on.

Weighted Fair•Packets placed into different queues according to some criteria (type, source…)•Service queue i for xi fraction of the time whereand T = total number of queues.

T

iix

0

1

Thomas Shen, University of Illinois-UC and Dr. K.C. Wang, Clemson University

Quad Chain

Pitfalls

TCP throughput for multi-hop traffic on the quad chain and small mesh were terrible

Lack of MAC access prevents packets from being sent

With few packets, queuing method has no effect

IEEE 802.11 MAC protocol is not efficient for multi-hop networks as documented in literature

Thresholds T1 and T2 were both varied. T2 controls Flow 1 and 2. At T2 = 0.5 , UDP flows 1 and 2 got the same end-to-end rate.

Conclusion•Results show throughput is unbalanced using FIFO•Priority queuing allocates bandwidth among flows•In our simulations, thresholds of 0.5 to 0.7 distributed throughput most equally

Figure2

Figure4

Our Priority Strategy

Queue 0

Queue 1

Queue 2

Routing Packets Own Packets Others’ Packets

Queue 0

Queue 1

Queue 2

If packets exist

MAC layer

If packets exist

If packets exist

else

Probability p

Probability 1-p

•Use a combination of strict priority and weighted fair queuing

•Categorize packets based on packet type and source.

•Service routing packets first, since routes needs to be established before other packets can reach their intended destination

For example, the packet assignment for the triple chain UDP scenario is shown in Figure4.

33

2

12

0

RO

UT

ING

RO

UT

ING

RO

UT

INGQ1 Q2 Q1 Q2 Q1 Q2

Figure3

Figure8

I

T

0

1

3

2

Flows

Triple Chain

Figure5

Figure6 Figure7

I

2

2

4 5

0 1 3

Small MeshFigure10

Figure11

In a wireless mesh network, routers are connected wirelessly as shown in Figure 1.

Traffic from end-users travel through different routes as shown above. The different lengths and contention along each route affects performance of the flows. Routes that service multiple flows are places where prioritization strategies can be utilized to adjust performance.

Internet

Figure1

Threshold change allocated bandwidth between Flow 2,4 and 5. Other flows were unaffected due to lack of serious contention.

Figure9

UDP flows are 200Kbps CBR traffic

Threshold

Threshold 1 Threshold 2

UDP flows are 200Kbps CBR traffic

UDP flows are 100Kbps CBR traffic

Threshold changed for all intermediate nodes.

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Triple Chain UDP ResultsFlow 0 (1 hop)Flow 1 (1 hop)Flow 2 (2 hop)Flow 3 (2 hop)Flow 0 OriginalFlow 1 OriginalFlow 2 OriginalFlow 3 Original

UD

P E

nd-t

o-e

nd S

ucc

ess

Rat

e

Threshold

0

50

100

150

200

250

300

350

0 0.2 0.4 0.6 0.8 1

Triple Chain TCP Results

Flow 0 (1 hop)Flow 1 (1 hop)Flow 2 (2 hop)Flow 3 (2 hop)Flow 0 OriginalFlow 1 OriginalFlow 2 OriginalFlow 3 Original

TC

P E

nd-t

o-en

d T

hro

ughp

ut (

Kb

ps)

Threshold

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Small Mesh UDP Results Flow 0 (1 hop)Flow 1 (2 hop)Flow 2 (2 hop)Flow 3 (2 hop)Flow 4 (3 hop)Flow 5 (3 hop)Flow 2 OriginalFlow 4 OriginalFlow 5 Original

UD

P E

nd-t

o-e

nd S

ucc

ess

Rat

e

Threshold