Wireless mesh network testbed measurements method and its accuracy

1
Wireless mesh network testbedmeasurements method
and its accuracy

• Networked Media Lab.
• Ubiquitous Media Team
• Jaeyong Yoo
• 25/4/2008

2
Wireless mesh networks

• Goal understand their properties and behavior
• Throughput, delay (single and multiple hops)
• TCP behavior

3
Challenges

• Basic requirement (and goal of this internship)
• Network measurement system with packet level
  accuracy
• Challenges
• Hardware limitations router performance
• Software limitations driver information
• Physical limitations causes of packet errors
• Correlation of information
• Monitors at different stations
• Different layers

4
Overall plan and current stage
Testbed construction and measurements
Trace processing and analysis
Network Observation
Parameter analysis
Wireless router
Wireless router
Traffic generator
Traffic receiver
Monitor center
Automatic measurement operator
Trace file processing
5
Wireless mesh network testbed setup

• Packet capturing by tcpdump in router

6
Network monitoring methods
Periodical observe
Packet embedded
Packet-arrival initiated
Network
Network
Unique ID
arbitrary precision - hard to correlate to
packet trace high overhead with high
precision,
packet-level accuracy - additional trace
gathering
low router overhead packet-level accuracy -
hard to apply in real-user scen.
7
How to collect network information?

• Consider an example of collecting network
  information while sending a packet

Router 1
Router 2
Sender
Receiver
8
How to collect information? (contd)
Sender protocol stack
Application
Router 1 protocol stack
TCP
timestamp
IP
timestamp
IP
ethernet
802.11
ethernet
Router 2
Router 1
Sender
Receiver
9
How to collect information? (contd)
Receiver protocol stack
Application
Router 2 protocol stack
TCP
Netfilter hook
IP
timestamp
IP
timestamp
ethernet
802.11
ethernet
Router 1
Router 2

• We got raw traces

Sender
Receiver
10
trace processing
Compare
Router 1
Router 2

• We got correlated traces

11
Monitor-able parameters
From packet trace
From MAC layer
From TCP layer
12
Validation of measurement accuracy

• So far
• We developed the measurement system
• Now,
• Perform measurements
• Varying parameters
• Modulation rate 11Mbps, 1Mbps
• MAC layer interface queue size 20, 50
• Validate the accuracy
• Correlating parameters

13
Setup

• Testbed
• In Safari 5 wireless routers (Asus)
• Multihop with static routing
• Limit to 3 hops
• Traffic generator
• Two TCP flows
• Maximum receiver window size
• TCP Reno
• Measurement time 10 minutes
• Wireless setup
• Rts/cts off, no antenna diversity
• Fixed bssid and channel 1
• Pseudo adhoc mode, 802.11g
• tx_queuelen 1

14
Case 1. Rate1 Mbps, queue limit50
Average throughput 268Kbps
Packet loss Due to queue drop (xtime) (yper
centage)
100
Router 2
0
100
Congestion window size (xtime) (ysegments)
80
60
40
Flow 1 Flow 2
20
0
3
Smoothed RTT (xtime) (yseconds)
2
1
Flow 1 Flow 2
0

• Large round trip time

15
Case 2. Rate1Mbps, queue limit20
Average throughput 265Kbps
Packet loss Due to queue drop (xtime) (yper
centage)
100
Router 1 Router 2 Router 3
0
40
Congestion window size (xtime) (ysegments)
20
Flow 1 Flow 2
0
Smoothed RTT (xtime) (yseconds)
1.5
1
0.5
Flow 1 Flow 2

• Dynamic congestion window changedue to smaller
  queue size

0
16
Case 2. Rate1Mbps, queue limit20 (Zoom in)
Average throughput 265Kbps
Packet loss Due to queue drop (xtime) (yper
centage)
100
Router 1 Router 2 Router 3
0
40
Congestion window size (xtime) (ysegments)
20
Flow 1 Flow 2
0
1.5
Smoothed RTT (xtime) (yseconds)
1
0.5
Flow 1 Flow 2
0
17
Case 3. Rate11Mbps, queue limit20
Average throughput 1813 Kbps
Packet loss At MAC/PHY layer (xtime) (yperc
entage)
100
Router 1 Router 2 Router 3
0
40
Congestion window size (xtime) (ysegments)
20
Flow 1 Flow 2
0
0.3
Smoothed RTT (xtime) (yseconds)
0.2
0.1
Flow 1 Flow 2

• No queue drop packet loss due tofrequent
  MAC/PHY layer packet loss

0
18
Case 3. Rate11Mbps, queue limit20 (Zoom in)
Packet loss at MAC/PHY layer (xtime) (yperc
entage)
100
Router 1 Router 2 Router 3
0
40
Congestion window size (xtime) (ysegments)
20
Flow 1
0
0.3
Smoothed RTT (xtime) (yseconds)
0.2
0.1
Flow 1
0
19
Intermediate results

• What have we achieved?
• Correlation between packet loss and congestion
  window size
• Now, more detailed correlation
• Queue length, retransmission counts, and RTT

20
Case 1. Rate1 Mbps, queue limit50
3
Analytical RTT (xtime) (yseconds)
2
1
3
Measured RTT (xtime) (yseconds)
2
1
0.7
Cross-correlation (xtime) (ycorrelation)
0

• Similarities btw. analytical and measured RTT

21
Conclusion

• Measurement system
• One-click measurement system
• One-click trace process/analysis
• Packet-level accuracy validated
• Open questions
• What is the cause of retransmissions?
• Future direction
• Applying real wireless mesh network
• Experimental evaluation of theoretical throughput
  models

22
Upgrade (done)
23
Performance (Synchronization)
24
Performance (Router overhead)CPU usage
25
Performance (Router overhead)Throughput
26
Performance (Router overhead)Delay
27
Future Plan - Objective

• Find out the reason of retransmission in real WMN
  testbed
• Retransmissions
• What are the possible reasons
• Collision
• Pure channel loss (due to weak signal)
• How hard is the detecting of reasons?
• Even device can not distinguish collision and
  channel error
• Possible Application
• As a supporting tool
• Analysis/validate Bianchi model in real testbed
• Evaluate Modulate Selection Algorithms
• Help other researchers to understand what is
  going on their testbed

28
Future Plan - Approach
Collision Monitoring System
Synchronized Trace files
Node A
Node B
Time line
Output
Node A
Node B
Synchronization
29
Backup slides
30
Hardware limitation of packet-level precision MAC
info (extend it to further case), correlate with
capture time point (cause burstiness of queue
length info)
Which parameters queue length, retransmission
count, channel error counts
Affects the information of processed
packets (bursty behavior)
10ms
madwifi
athstats (MAC layer stats)
Soft-part
txtasklet
Few delay
HAL (hardware)
16 packets
Correct information
31
Measurement Automation

• Requirement Wired connection to the node

• Basically, four actions
• Network configuration
• Monitoring tools initiation
• Traffic generation
• Trace collection

32
Measurement topology and base line throughput
(1Mbps case)
901Kbps 0 loss (uni) 0.31 loss (broad)
899Kbps 0 loss(uni) 0.12 loss (broad)
897Kbps 0 loss (uni) 0.14 loss(broad)

• End-to-end throughput
• (UDP, sending rate 350Kbps) 295Kbps

33
Measurement topology and base line throughput
(11Mbps case)
5.6Mbps 0.19 loss (uni) 4.12 loss (broad)
6.8 Kbps 0 loss (uni) 0.48 loss(broad)
6.5Mbps 0 loss(uni) 0.3 (broad)

• End-to-end throughput
• (UDP, sending rate 2.2Mbps) 2.09Mbps

34
Nominal relation between queue occu. length,
MAC layer retran., and RTT
Packet of our interest
How long will it take to be delivered to the next
router ? Queuing delay Transmission time
(Queue length) (Retransmission Count)
Queue length
Retransmissions
35
RTT calculation
TCP sender
TCP receiver
Router 1
Router 2
Router n

• Assume half of queue filled by TCP data packet
  and half of queue filled by TCP ack packet
• Average transmission time of data/ack packet Td
  , Ta
• qi (t) Queue length of router i at time t (from
  measurement)
• ri (t) Average retransmission of router i at
  time t (from measurement)
• ti (t) Number of transmissions to be done at
  router i at time t
• ti (t) n I
• Average number of total transmissions of data/ack
  packet at time t
• Ad (t) ? ri (t) x ti (t) x qi (t) / 2
• Aa (t) ? ri (t) x (i 1) x qi (t) / 2
• Expected RTT at time t
• R(t) Ad(t)xTd Aa(t) x Ta

36
Monitoring system validation (1Mbps case, queue
limit 20)Queue occupation length, RTT (0150
seconds)
Estimated RTT 0.8 sec
6 packets
Queue Length Of 1st router
Queue Length Of 2nd router
10 packets
Queue Length Of 3rd router
6 packets
Measured RTT
1 sec
0.5 sec
37
Monitoring system validation (1Mbps case, queue
limit 50)Packet trace of individual router
(99102 seconds)
Data transmission
Router 1 data packet arrival
Blue Data packet of interest
Red Other packets
Router 2 data packet arrival
R1-gtR2
Router 3 data packet arrival
R2-gtR3
Router 4 data packet arrival
R3-gtR4
Blue Corresponding ACK packet
Router 1 ack packet arrival
Router 2 ack packet arrival
ACK transmission
R3-gtR2
Router 3 ack packet arrival
R2-gtR1
Router 4 ack packet arrival
38
Case 1. Rate 1 Mbps, queue limit 20
Analytical RTT (yseconds) (xtime)
1.5
1
0.5
0
Measured RTT (yseconds) (xtime)
1.5
1
0.5
0
0.7
Cross-correlation (ycorrelation) (xtime)
0
39
Case 1. Rate 11 Mbps, queue limit 20
Analytical RTT (ytime) (xtime)
0.4
0.3
0.2
0.1
Measured RTT (ytime) (xtime)
0.4
0.3
0.2
0.1
Cross-correlation (ycorrelation) (xtime)
0.8
0
40
MAC layer information reconstruction (1Mbps
case, queue limit 50) Packet trace of
individual router (100.1100.2 seconds)
Receiving ack packet
Sending out of data packet
41
MAC layer information reconstruction (1Mbps
case, queue limit 50) Packet trace of
individual router (343.85344.35 seconds)
5 receiving errors
3 retransmissions
collision or channel error ?
42
MAC layer information reconstruction (1Mbps
case, queue limit 50) What has to be improved?
exact time of retransmission ?
Also, we have to distinguish reason of
retransmission (Collision? Or Channel error?)