Broadband Wireless Local Loop
Joseph B. Evans
Charles E. Spahr Professor of EECS
Acting Director, ITTC
University of Kansas
evans@ittc.ukans.edu
Sprint Research Symposium 2000
University of Kansas
Outline
•
•
•
•
•
•
Motivation
Performance experiments
Long-term monitoring
MAC performance
Spectrum issues
Conclusions
University of Kansas
Motivation
• Determine suitability of wireless for local access
• Components
• assess limits of technology via experimentation
– performance (throughput and delay)
– long-term monitoring (reliability)
• media access layer performance
– compare different approaches to MAC layer
– assess capability of MAC layers to provide particular services
• spectrum issues
– working with under-utilized frequency bands
University of Kansas
Faculty and Students
• Faculty
• Joe Evans
• Jim Roberts
• Students
•
•
•
•
•
Mihir Thaker
Harish Sitaraman
Jesse Davis
Dragan Trajkov
Larry Sanders
• Sprint
• Manish Mangal (now at Sprint PCS)
University of Kansas
Infrastructure
University of Kansas
Experiment Infrastructure
• Prototype high bandwidth wireless solution to deliver
integrated data, voice, and video services to the home
or small business
RF Backhaul
Fiber Backhaul
Cell
Site
Wide Area Network
Internet
voice
video
data
University of Kansas
Experiment Infrastructure
• Example wireless modem
•
•
•
•
AB-Access from Adaptive Broadband
25 Mb/s shared per cellular sector in U-NII band
up to 256 users per sector
3 km to 5 km range
University of Kansas
Experiment Infrastructure
Staff
Member’s
Home
ITTC
Nichols Hall
University of Kansas
Performance Experiments
University of Kansas
Performance Tests
• Subscriber Unit (SU) #2 located 2.4 km from Access Point
(AP), SU #3 located 10 m from AP
• NetSpec used to conduct measurement tests
• per-host as well as aggregate bandwidth measured
• TCP streams used for tests
• Tests performed
SU Test Host
• flooding of link to determine maximum throughput
• contention tests
Router
SU modem
AP
controller
SU #2
AP modem
VoIP gear
SU #3
SU modem
Internet
SU Test Host
University of Kansas
Performance Test Results
• Since transmitting over
Ethernet from host to
modem on residential side,
throughput limited to less
than 10 Mb/s
• Fluctuations in SU #2
bandwidth most likely due
to TCP retransmits
7.7
7.65
Throughput (Mb/s)
• data transmitted towards and
measured on AP
7.75
7.6
8192
7.55
16384
7.5
32768
65536
7.45
7.4
7.35
7.3
4096
8192
16384
24576
32768
40960
49152
57344
65536
Message Size (B)
Throughput, Full Link, SU #3
9.15
9.1
Throughput (Mb/s)
• Link flooded to determine
link capacity with different
buffer sizes
Throughput, Full Link, SU #2
8192
9.05
16384
32768
9
65536
8.95
8.9
4096
8192
16384
24576
32768
40960
49152
57344
65536
Message Size (B)
University of Kansas
Performance Test Results
Aggregate Throughput, Contending Traffic
14
12
Throughput (Mb/s)
• Contention test
consisted of both SUs
transmitting towards AP
• Generally, SU #2
attained higher
bandwidth than SU #3
• Total bandwidth about 2
Mb/s slower than
predictions from vendor
10
8
16384
6
32768
4
2
0
4096
8192
16384 24576 32768 40960 49152 57344 65536
Message Size (B)
University of Kansas
Performance Test Results
Aggregate Throughput, Bidirectional Traffic
7.8
7.6
Throughput (Mb/s)
• Traffic sent from both
sides of link between AP
and SU #2
• Since transmitting over
Ethernet from host to
modem on residential
side, throughput limited
to less than 10 Mb/s
7.4
16384
7.2
32768
7
65536
6.8
6.6
6.4
4096
8192
16384
24576
32768
40960
49152
57344
65536
Message Size (B)
University of Kansas
Long-term Monitoring
University of Kansas
Long-term Monitoring - Motivation
• Customers want stable, fast, and low delay TCP
connections
• emerging always-on networks
• most traffic is TCP-based
• Software developed to measure, collect, and archive
measurements on TCP connections
• New tool called “ConMon”
University of Kansas
Data Collection
• Stability
• frequency of connection loss
• reason for the connection loss behavior
• Throughput
• variations and asymmetries
• Delays
• round trip time
• different packet sizes
University of Kansas
ConMon - Connection Monitor
Client 1
Data Collection Host
ConMon
Process
TCP connections
ConMon
Process
Traceroute
Process
ConMon
Traceroute
Traceroute
Process
Client 2
University of Kansas
ConMon Features
•
•
•
•
Continuous monitoring
Measures connection stability, throughput and RTT
Client reconnects to server when connection drops
Excessive congestion results in a connection loss
• from an end-user’s perspective, long idle times
• timers used to terminate a connection
• Traces the current path between the client & server
• provides method to discern the reason for a connection loss
• Uses a plotter and table generator to display results
University of Kansas
ConMon Operation
• Server runs on a central data collection machine
• Maintains information about parameters to use
• packet size, frequency of transmission, number of packets/sample etc...
•
•
•
•
Clients connect to the server
Server notifies traceroute daemon
Connection is monitored from both ends
Client or server performs throughput measurement
University of Kansas
ConMon Traceroute Facility
• Modified version of the popular traceroute
• Notified by ConMon daemon about client status
• starts monitoring the path when client connects
• stops after a timeout when a connection is lost
• Maintains a file containing number of hops, hop at
which route changed and time
• File updated on observation of route change or hop
unreachable
University of Kansas
Preliminary Results
• Did not have throughput and traceroute features
• Initial results - connection very unstable
• approximately 4-5 losses per day
• round trip times varied widely
• Problem rectified
• radio’s were operating on an indoor channel
• reconfigured both the SU and the AP for outdoor testing
University of Kansas
Preliminary Results
• Connection more stable after reconfiguration
• around 1-2 drops per day
• Simultaneous throughput tests using NetSpec
• connection drops increased due to link saturation
• loss rate increased as observed using mtr
• Tests also performed on cable modem network
University of Kansas
Preliminary ConMon Results
• Comparison for different months
Media
Drops
RTT
Throughput
(ms)
(Mb/s)
B-WLL
114
1.96
-
Cable
19
77
0.56
Drops
RTT
Throughput
(ms)
(Mb/s)
Media
B-WLL
32
1.96
-
Cable
10
98
0.437
University of Kansas
Media Access Layer
Performance
University of Kansas
MAC Layer Evaluation - Motivation
• Various MAC layer solutions offered by vendors
• Compare different approaches to MAC layer
• media shared amongst customers
• Assess capability of MAC layers to provide particular
services
• basic best-effort Internet services
• qualities of service or differentiated services
University of Kansas
MAC Layer Evaluation Strategy
• Model components of typical broadband wireless
MAC layers
• contention mode
• data transfer mode
• Simulate using various traffic types
• OPNET tool used
• built-in traffic models used
• particular MAC layers modeled and tested
University of Kansas
MAC Layer Simulation
• Model physically separated server and workstation
with MAC residing on the radio nodes
• TCP applications running end-to-end
University of Kansas
MAC Layer Simulation
• User application configuration
University of Kansas
MAC Layer Simulation
• HTTP traffic, many images
% of Max. Througput
Throughput Degradation
120
100
100
86.16
71.48
55.35
80
60
40
48.9844.3
40.71
20
0
0
2
4
6
# of Users/channel
8
10
University of Kansas
MAC Layer Evaluation Strategy
• Queuing delays due to channel contention
Delay Characteristics
% increase over
lowest value
300
273.68
250
226.31
182.45
142.1
200
150
100
91.22
47.36
50
0
0
0
1
2
3 4 5 6 7
# users/channel
8
9 10
University of Kansas
MAC Layer Evaluation Strategy
• Throughput results
Normalized
throughput
TDMA Throughput
1.3
1.25
1.2
1.15
1.1
1.05
1
0.95
0.9
0
1
2
3
4
5
6
7
8
9
10 11
# Users/fram e
University of Kansas
MAC Layer Evaluation Comments
• Compare on the basis of common metrics such as delay x BW,
throughput x number of users
• TDMA based systems more efficient for QoS delivery than
MF-Polling
• TDMA scheduler can be redesigned to adapt to the existing
traffic and larger user population
• MF-Polling can be effectively improved and changed to MFTDMA for supporting QoS based applications
• MF-TDMA schemes are most effective for scarce (!!)
bandwidth with good support for intensive applications
University of Kansas
Spectrum Issues
University of Kansas
Spectrum Issues - Motivation
• Determine if it is possible to use a previously
allocated frequency spectrum in a way that would not
cause harmful interference to all other users in the
area that use the same frequency
• Create a tool which will locate an adequate position
for the access point in such manner that it does not
cause interference, given that the data for the other
antennas is provided
University of Kansas
Interference Tool
• Receiver data required (for each receiver)
•
•
•
•
type of antenna (circular or rectangular aperture)
elevation angle
azimuth angle
x, y and z coordinate (z represents height of antenna)
• Transmitter data required
•
•
•
•
•
•
•
type of antenna
elevation angle
azimuth angle
radiating power
frequency
dimensions of the antenna
z coordinate (height only)
University of Kansas
Interference Tool Operation
• Positions the transmitter on a certain location (x2,y2) and calculates the gains
of the transmitter and the receiver in the direction of one another
• Using the two ray model, and desired interference level, calculates whether
there is interference or not
• Then moves transmitter
to another location
(x3,y3) and repeats
• Repeated for entire area
of interest in the xy plane
• Procedure is repeated for
each receiver
(x1,y1,z1)
(x3,y3,z3)
z3=z2
x3
(x2,y2,z2)
y3
University of Kansas
Simulation Example
• One AP and one satellite receiver in the area
• question - how close can the AP be placed, given that in the azimuth
plane it always looks into the receiver?
• two simulation results, for different antenna heights
• Assumptions
• AP antenna type is rectangular aperture
• antenna type of the other users is circular aperture
• harmful interference level is 1 dB
University of Kansas
Simulation Example
AP antenna height at 2 m
AP antenna height at 30 m
University of Kansas
Conclusion
• Broadband wireless is a complex environment
• service, link, and physical layer considerations
• Studying environment at different layers to insure that
reliable and high performance services can be
delivered
University of Kansas