Professional Documents
Culture Documents
Wireless and Mobile Computing Project Presentation EE 382V project Sunil Kowlgi Vacha Dave
Presentation Outline
Motivation
Most wireless cards come with multiple data rates The commonly held belief is that higher data rate guarantees higher throughput. Higher data rates require higher transmit power which means lesser spatial reuse.
We would like to see how spatial reuse is affected for different data rates and how this effect manifests as reduced throughput.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 3
Methodology Overview
We measured throughput for a single flow on a linear chain topology of nodes. Measurements were carried out for different data rates and transmit power levels. The number of concurrent successful transmissions was measured to serve as an indication of spatial reuse.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 4
Phy/WirelessPhy:
Frequency 2.4GHz (802.11b) RX threshold varied according to data rate Transmit power swept over a range 1 30dBm Carrier sense threshold - 1e-13W Capture threshold 10dB
Propagation/Shadowing:
Path Loss Exponent 2.0 (free space) Shadowing deviation 4.0 ( outdoor)
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 5
Mac/802_11 :
Package that enables static routing for wireless nodes Lets you set the routing table by hand for each node Queue/DropTail/PriQueue interface queue type Network interface queue length 500
Constant Bit Rate (CBR) traffic is used CBR parameters that were set:
Time interval between packets varied for different experiments Packet size 1456 bytes + 24 MAC hdr + 20 IP hdr
Different transmit data rates can be achieved by varying CBR parameters CBR lets you limit the maximum number of packets that need to be sent from source to sink.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 8
FCC mandates that wireless transmitters limit power to 30dBm (1W). Commonly used transmit power level for wireless cards is 15dBm. We ran experiments for all power levels in the range 1 30dBm.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 9
Loss Monitor was used to measure the throughput at the sink, over a given time window. All simulations were run for 700 seconds The initial and final 100 seconds of simulation were excluded from throughput measurement. This is to ensure that the system was in steady state.
10
To determine the chain length To determine of CBR value To measure the throughput for different power levels, for 802.11b data rates Number of concurrent transmissions for different power levels, for 802.11b data rates
11
Trying to approximate an infinite chain length, for which throughput does not vary significantly with small change in number of nodes. Before carrying out throughput v. data rate experiments we determined the right chain length and that would be valid for different data rates. Settings:
Power level 15dBm Packet size 1500 bytes Distance between nodes 550m
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 12
1000 900 800 700 600 500 400 300 200 100 0 0 10 20 30 40 50 Chain Length 60 70 80 90
Throughput (Kbps)
13
Throughput in Kbps
14
Throughput (Kbps)
15
There is a quasi-exponential decay in throughput as the chain length is increased. The chain length for our experiments was picked from the stable/flat region of the throughput curve.
We need to determine the CBR value at which there are no congestion losses. For a chain length of 60 nodes, we measured the throughput for different CBR values. This was carried out for 1, 2 and 11Mbps data rates, at a transmit power level of 15dBm.
17
Throughput (Kbps)
18
Throughput (Kbps)
1000
10000
19
20
Throughput increases with CBR value until a point, where it starts to fall. The reduction in throughput beyond a certain CBR value can be attributed to congestion. For our subsequent experiments we use CBR values just below the inflection point so that it guarantees reasonable throughput without the worry of congestion. For 1Mbps 31.6 Kbps For 2Mbps 60 Kbps For 11Mbps 60 Kbps
21
Simulations were run to measure the throughput for 30 different transmit power levels ( 1- 30 dBm). These simulations were performed for different data rates 1, 2, 11 Mbps. CBR values determined in the previous experiment were used ( to avoid congestion-related losses). Nodes were separated by 550m.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 22
15 dBm
Throughput(Kbps)
0.1
23
For very low values of transmit power, very few data packets get across from source to sink, and throughput is very low. As the transmit power increases, transmissions become more reliable and the throughput increases till a point. There is an inflection point corresponding to a certain transmit power beyond which fewer nodes transmit in a given time interval and thus spatial reuse decreases.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 24
Power v. Throughput for CBR=31.6Kbps, chain length = 60, 550m between nodes
35 30 15 dBm
Power v. Throughput for CBR=60Kbps, chain length = 60, 550m between nodes
70 60
15 dBm
Throughput(Kbps)
Throughput(Kbps)
0.1
0.1
25
For each data rate, throughput increases with increased transmit power and beyond a certain transmit power it starts to fall. The graphs are testament to the fact that for a given transmit power a higher data rate does not always result in higher throughput. The envelope of the curve gives the data rate at which transmissions should happen, to achieve the highest throughput possible for a given transmit power.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 26
The 11Mbps throughput-transmit power curve appears to flatten out, contrary to intuition! But, if the curve is extrapolated for transmit powers up to 5W (!) it shows the degradation of throughput.
27
The minimum distance between nodes at which an optimal Fairness Ratio of 1 is achieved is indicative of spatial reuse. The number of transmissions that complete in a time interval ( defined by the transmission time of a standard size data packet) is indicative of spatial reuse. Using the queuing theory
First packet 59x , second at 59x + x, and so on Measuring the deviation from ideal
After a simulation run, the trace file was examined by a script to determine the maximum number of ACKs in a time interval. The time interval was determined by measuring time from sending of RTS to the reception of an ACK, for a single CBR packet of 1500 bytes.
29
0.1
Power v. Throughput for CBR=60Kbps, chain length = 60, 550m between nodes
70 60
15 dBm
Throughput(Kbps)
0.1
30
Very Jittered Curve seems to follow the general throughput curve Binning the values and taking a weighted average would result in a better curve Pessimistic bound since it counts only the transmissions that have completed during the given time interval.
31
We have maintained a distance on 550 m between nodes. This is just under the transmission range for 1Mbps.
More exhaustive sweep over different Application layer data rates. Trying out for different packet sizes
Throughput (Kbps)
30
33
Issues
We could not see any throughput for 5.5 Mbps, for all transmit power levels. Possibly because of incorrect value of RX threshold. The default TTL value for IP in NS-2 is set to 32. A week was spent in figuring out the mysterious packet drops for long chain lengths. NS-2 simulations took a long time, which prevented us from performing more exhaustive experiments. For instance, sweeping across different CBR rates for different power levels.
Sunil Kowlgi, Vacha Dave Univesity of Texas at Austin 34
References
The Network Simulator http://www.isi.edu/nsnam/ns/ Capacity of Ad Hoc Wireless Networks Jinyang Li, Charles Blake, Douglas S. J. De Couto, Hu Imm Lee, and Robert Morris , MOBICOM 01 Improving spatial reuse through tuning transmit power, carrier sense threshold, and data rate in multihop wireless networks, Kim et al. MOBICOM 2006
An Experimental Evaluation of Several Rate Adaptation Protocols, Choi et. al.
35