28-05-2012, 11:23 AM
Buffer Sizing for 802.11-Based Networks
[attachment=23269]
Abstract
We consider the sizing of network buffers in IEEE
802.11-based networks.Wireless networks face a number of fundamental
issues that do not arise in wired networks.We demonstrate
that the use of fixed-size buffers in 802.11 networks inevitably
leads to either undesirable channel underutilization or unnecessary
high delays. We present two novel dynamic buffer-sizing
algorithms that achieve high throughput while maintaining low
delay across a wide range of network conditions. Experimental
measurements demonstrate the utility of the proposed algorithms
in a production WLAN and a lab test bed.
Index Terms—Buffer sizing, IEEE 802.11, IEEE 802.11e,
medium access control (MAC), stability analysis, Transmission
Control Protocol (TCP), wireless LANs (WLANs).
I. INTRODUCTION
I N COMMUNICATION networks, buffers are used to
accommodate short-term packet bursts so as to mitigate
packet drops and to maintain high link efficiency. Packets are
queued if too many packets arrive in a sufficiently short interval
of time during which a network device lacks the capacity to
process all of them immediately.
For wired routers, the sizing of buffers is an active research
topic [5], [9], [27], [31], [32]. The classical rule of thumb for
sizing wired buffers is to set buffer sizes to be the product of
the bandwidth and the average delay of the flows utilizing this
link, namely the Bandwidth-Delay Product (BDP) rule [31]. See
Section VII for a discussion of other related work.
Surprisingly, however, the sizing of buffers in wireless
networks (especially those based on IEEE 802.11/802.11e)
appears to have received very little attention within the networking
community. Exceptions include the recent work in
[21] relating to buffer sizing for voice traffic in 802.11e [2]
wireless LANs (WLANs), work in [23] that considers the impact
of buffer sizing on Transmission Control Protocol (TCP)
upload/download fairness, and work in [29] that is related to
802.11e parameter settings.
Buffers play a key role in 802.11/802.11e wireless networks.
To illustrate this, we present measurements from the production
WLAN of the Hamilton Institute, Maynooth, Ireland, which
show that the current state of the art, which makes use of
fixed-size buffers, can easily lead to poor performance. See
Manuscript received August 19, 2008; revised August 07, 2009 and April
22, 2010; accepted June 20, 2010; approved by IEEE/ACM TRANSACTIONS ON
NETWORKING Editor M. Buddhikot. Date of publication November 18, 2010;
date of current version February 18, 2011. This work was supported by the Irish
Research Council for Science, Engineering and Technology and Science Foundation
Ireland Grant 07/IN.1/I901.
The authors are with the Hamilton Institute, National University of Ireland
Maynooth, Maynooth, Ireland (e-mail: tianji.li[at]nuim.ie; doug.leith[at]nuim.ie;
david.malone[at]nuim.ie).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNET.2010.2089992
the Appendix for further details of the configuration used.
We recorded round-trip times (RTTs) before and after one
wireless station started to download a 37-MB file from a Web
site. Before starting the download, we pinged the access point
(AP) from a laptop five times, each time sending 100 ping
packets. The RTTs reported by the ping program were between
2.6–3.2 ms. However, after starting the download and allowing
it to continue for a while (to let the congestion control algorithm
of TCP probe for the available bandwidth), the RTTs to the
AP increased by three orders of magnitude to 2900–3400 ms.
During the test, normal services such as Web browsing experienced
obvious pauses/lags on wireless stations using the
network. Closer inspection revealed that the buffer occupancy
at the AP exceeded 200 packets most of the time and reached
250 packets from time to time during the test. Note that the
increase in measured RTT could be almost entirely attributed
to the resulting queuing delay at the AP and indicates that
a more sophisticated approach to buffer sizing is required.
Indeed, using the algorithm proposed in this paper, the
RTTs observed when repeating the same experiment fall to
only 90–130 ms. This reduction in delay does not come at the
cost of reduced throughput; i.e., the measured throughput with
the algorithm and the default buffers is similar.
In this paper, we consider the sizing of buffers in
802.11/802.11e-based [1], [2] WLANs.We focus on single-hop
WLANs since these are rapidly becoming ubiquitous as the last
hop on home and office networks as well as in so-called “hot
spots” in airports and hotels, but note that the proposed schemes
can be easily applied in multihop wireless networks. Our main
focus in this paper is on TCP traffic since this continues to
constitute the bulk of traffic in modern networks (80%–90%
[35] of current Internet traffic and also WLAN traffic [28]),
although we extend consideration to UDP traffic at various
points during the discussion and also during our experimental
tests.
Compared to sizing buffers in wired routers, a number of
fundamental new issues arise when considering 802.11-based
networks. First, unlike wired networks, wireless transmissions
are inherently broadcast in nature, which leads to the packet
service times at different stations in a WLAN being strongly
coupled. For example, the basic 802.11 DCF ensures that the
wireless stations in a WLAN win a roughly equal number of
transmission opportunities [19], hence the mean packet service
time at a station is an order of magnitude longer when 10 other
stations are active than when only a single station is active.
Consequently, the buffering requirements at each station would
also differ, depending on the number of other active stations
in the WLAN. In addition to variations in the mean service
time, the distribution of packet service times is also strongly dependent
on the WLAN offered load. This directly affects the
burstiness of transmissions and hence buffering requirements
(see Section III for details). Second, wireless stations dynamically
adjust the physical transmission rate/modulation used in
1063-6692/$26.00 © 2010 IEEE
LI et al.: BUFFER SIZING FOR 802.11-BASED NETWORKS 157
order to regulate noncongestive channel losses. This rate adaptation,
whereby the transmit rate may change by a factor of 50
or more (e.g. from 1 to 54 Mb/s in 802.11a/g), may induce large
and rapid variations in required buffer sizes. Third, the ongoing
802.11n standards process proposes to improve throughput efficiency
by the use of large frames formed by aggregation of
multiple packets [3], [18]. This acts to couple throughput efficiency
and buffer sizing in a new way since the latter directly
affects the availability of sufficient packets for aggregation into
large frames.
It follows from these observations that, among other things,
there does not exist a fixed buffer size that can be used for sizing
buffers in WLANs. This leads naturally to consideration of dynamic
buffer-sizing strategies that adapt to changing conditions.
In this paper, we demonstrate the major performance costs associated
with the use of fixed buffer sizes in 802.11 WLANs
(Section III) and present two novel dynamic buffer-sizing algorithms
(Sections IV and V) that achieve significant performance
gains. The stability of the feedback loop induced by
the adaptation is analyzed, including when cascaded with the
feedback loop created by TCP congestion control action. The
proposed dynamic buffer-sizing algorithms are computationally
cheap and suited to implementation on standard hardware. Indeed,
we have implemented the algorithms in both the NS-2
simulator and the Linux MadWifi driver [4]. In this paper, in
addition to extensive simulation results, we also present experimental
measurements demonstrating the utility of the proposed
algorithms in a test bed located in office environment and
with realistic traffic. The latter includes a mix of TCP and UDP
traffic, a mix of uploads and downloads, and a mix of connection
sizes.
The remainder of the paper is organized as follows. Section II
introduces the background of this work. In Section III, simulation
results with fixed-size buffers are reported to further
motivate this work. The proposed algorithms are then detailed
in Sections IV and V. Experiment details are presented in
Section VI. After introducing related work in Section VII, we
summarize our conclusions in Section VIII.
II. PRELIMINARIES
A. IEEE 802.11 DCF
IEEE 802.11a/b/g WLANs all share a common MAC algorithm
called the Distributed Coordinated Function (DCF),
which is a CSMA/CA-based algorithm. On detecting the
wireless medium to be idle for a period , each wireless
station initializes a backoff counter to a random number selected
uniformly from the interval [0, CW-1] where CW is the
contention window. Time is slotted and the backoff counter is
decremented each slot that the medium is idle. An important
feature is that the countdown halts when the medium is detected
busy and only resumes after the medium is idle again for a
period . On the counter reaching zero, a station transmits
a packet. If a collision occurs (two or more stations transmit
simultaneously), CW is doubled and the process repeated. On
a successful transmission, CW is reset to the value and
a new countdown starts.
B. IEEE 802.11e EDCA
The 802.11e standard extends the DCF algorithm (yielding
the EDCA) by allowing the adjustment of MAC parameters that
Fig. 1. WLAN topology used in simulations. Wired backhaul link bandwidth
100 Mb/s. MAC parameters of the WLAN are listed in Table I.
TABLE I
MAC/PHY PARAMETERS USED IN SIMULATIONS, CORRESPONDING TO 802.11g
were previously fixed. In particular, the values of (called
in 802.11e) and may be set on a per-class basis
for each station. While the full 802.11e standard is not implemented
in current commodity hardware, the EDCA extensions
have been widely implemented for some years.
C. Unfairness Among TCP Flow
Consider a WLAN consisting of client stations each carrying
one TCP upload flow. The TCP ACKs are transmitted by
the wireless AP. In this case, TCP ACK packets can be easily
queued/dropped due to the fact that the basic 802.11 DCF ensures
that stations win a roughly equal number of transmission
opportunities. Namely, while the data packets for the flows
have an aggregate share of the transmission opportunities,
the TCPACKs for the flows have only a share.
Issues of this sort are known to lead to significant unfairness
among TCP flows, but can be readily resolved using 802.11e
functionality by treating TCP ACKs as a separate traffic class
that is assigned higher priority [15]. With regard to throughput
efficiency, the algorithms in this paper perform similarly when
the DCF is used and when TCP ACKs are prioritized using the
EDCA as in [15]. Per-flow behavior does, of course, differ due
to the inherent unfairness in the DCF, and we therefore mainly
present results using the EDCA to avoid flow-level unfairness.
D. Simulation Topology
In Sections III, IV, and V-G, we use the simulation topology
shown in Fig. 1, where the AP acts as a wireless router between
the WLAN and the Internet. Upload flows originate from stations
in the WLAN on the left and are destined to wired host(s)
in the wired network on the right. Download flows are from
the wired host(s) to stations in the WLAN. We ignore differences
in wired bandwidth and delay from the AP to the wired
hosts, which can cause TCP unfairness issues on the wired side
(an orthogonal issue) by using the same wired-part RTT for all
flows. Unless otherwise stated, we use the IEEE 802.11g PHY
parameters shown in Table I and the wired backhaul link bandwidth
is 100 Mb/s with RTT 200 ms. For TCP traffic, the widely
deployed TCP Reno with SACK extension is used. The advertised
window size is set to be 4096 packets (each has a payload
158 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 19, NO. 1, FEBRUARY 2011
Fig. 2. Measured distribution of per-packet MAC service time. Solid vertical
lines mark the mean values of distributions. Physical layer data/basic rates are
11/1 Mb/s. (a) 2 stations. (b) 12 stations.
of 1000 bytes), which is the default size of current Linux kernels.
The maximum value of the TCP smoothed RTT measurements
(sRTT) is used as the measure of the delay experienced
by a flow.
III. MOTIVATION AND OBJECTIVES
Wireless communication in 802.11 networks is time-varying
in nature; i.e., the mean service time and the distribution of
service time at a wireless station vary in time. The variations are
primarily due to: 1) changes in the number of active wireless
stations and their load (i.e., offered load on the WLAN); and
2) changes in the physical transmit rate used (i.e., in response
to changing radio channel conditions). In the latter case, it
is straightforward to see that the service time can be easily
increased/decreased using low/high physical layer rates. To see
the impact of offered load on the service time at a station, Fig. 2
plots the experimentally measured distribution of the MAC
layer service time when there are two and 12 stations active.
It can be seen that the mean service time changes by over an
order of magnitude as the number of stations varies. Observe
also from these measured distributions that there are significant
fluctuations in the service time for a given fixed load. This is a
direct consequence of the stochastic nature of the CSMA/CA
contention mechanism used by the 802.11/802.11e MAC.
This time-varying nature directly affects buffering requirements.
Fig. 3 plots link utilization1 and max sRTT (propagation
plus smoothed queuing delay, with smoothing via the standard
TCP RTT filter) versus buffer size for a range of WLAN offered
loads and physical transmit rates.We can make a number of observations.
First, it can be seen that as the physical-layer transmit rate
is varied from 1 to 216 Mb/s, the minimum buffer size to ensure
at least 90% throughput efficiency varies from about 20 to
about 800 packets. No compromise buffer size exists that ensures
both high efficiency and low delay across this range of
transmit rates. For example, a buffer size of 80 packets leads to
RTTs exceeding 500 ms (even when only a single station is active,
and so there are no competing wireless stations) at 1 Mb/s
and throughput efficiency below 50% at 216 Mb/s. Note that
the transmit rates in currently available draft 802.11n equipment
already exceed 216 Mb/s (e.g., 300 Mb/s is supported by
current Atheros chipsets) and the trend is toward still higher
transmit rates. Even across the restricted range of transmit rates
1Here, the AP throughput percentage is the ratio between the actual
throughput achieved using the buffer sizes show on the x-axis and the maximum
throughput over the range of buffer sizes shown on the x-axis.
1–54 Mb/s supported by 802.11a/b/g, it can be seen that a buffer
size of 50 packets is required to ensure throughput efficiency
above 80%, yet this buffer size induces delays exceeding 1000
and 3000 ms at transmit rates of 11 and 1 Mb/s, respectively.
Second, delay is strongly dependent on the traffic load and
the physical rates. For example, as the number of competing
stations (marked as “uploads” in the figure) is varied from 0 to
10, for a buffer size of 20 packets and physical transmit rate
of 1 Mb/s, the delay varies from 300 to over 2000 ms. This
reflects that the 802.11 MAC allocates available transmission
opportunities equally on average among the wireless stations,
and so the mean service time (and thus delay) increases with the
number of stations. In contrast, at 216 Mb/s, the delay remains
below 500 ms for buffer sizes up to 1600 packets.
Our key conclusion from these observations is that there
exists no fixed buffer size capable of ensuring both high
throughput efficiency and reasonable delay across the range
of physical rates and offered loads experienced by modern
WLANs. Any fixed choice of buffer size necessarily carries
the cost of significantly reduced throughput efficiency and/or
excessive queuing delays.
This, therefore, naturally leads to the consideration of adaptive
approaches to buffer sizing, which dynamically adjust the
buffer size in response to changing network conditions to ensure
both high utilization of the wireless link while avoiding unnecessarily
long queuing delays.
[attachment=23269]
Abstract
We consider the sizing of network buffers in IEEE
802.11-based networks.Wireless networks face a number of fundamental
issues that do not arise in wired networks.We demonstrate
that the use of fixed-size buffers in 802.11 networks inevitably
leads to either undesirable channel underutilization or unnecessary
high delays. We present two novel dynamic buffer-sizing
algorithms that achieve high throughput while maintaining low
delay across a wide range of network conditions. Experimental
measurements demonstrate the utility of the proposed algorithms
in a production WLAN and a lab test bed.
Index Terms—Buffer sizing, IEEE 802.11, IEEE 802.11e,
medium access control (MAC), stability analysis, Transmission
Control Protocol (TCP), wireless LANs (WLANs).
I. INTRODUCTION
I N COMMUNICATION networks, buffers are used to
accommodate short-term packet bursts so as to mitigate
packet drops and to maintain high link efficiency. Packets are
queued if too many packets arrive in a sufficiently short interval
of time during which a network device lacks the capacity to
process all of them immediately.
For wired routers, the sizing of buffers is an active research
topic [5], [9], [27], [31], [32]. The classical rule of thumb for
sizing wired buffers is to set buffer sizes to be the product of
the bandwidth and the average delay of the flows utilizing this
link, namely the Bandwidth-Delay Product (BDP) rule [31]. See
Section VII for a discussion of other related work.
Surprisingly, however, the sizing of buffers in wireless
networks (especially those based on IEEE 802.11/802.11e)
appears to have received very little attention within the networking
community. Exceptions include the recent work in
[21] relating to buffer sizing for voice traffic in 802.11e [2]
wireless LANs (WLANs), work in [23] that considers the impact
of buffer sizing on Transmission Control Protocol (TCP)
upload/download fairness, and work in [29] that is related to
802.11e parameter settings.
Buffers play a key role in 802.11/802.11e wireless networks.
To illustrate this, we present measurements from the production
WLAN of the Hamilton Institute, Maynooth, Ireland, which
show that the current state of the art, which makes use of
fixed-size buffers, can easily lead to poor performance. See
Manuscript received August 19, 2008; revised August 07, 2009 and April
22, 2010; accepted June 20, 2010; approved by IEEE/ACM TRANSACTIONS ON
NETWORKING Editor M. Buddhikot. Date of publication November 18, 2010;
date of current version February 18, 2011. This work was supported by the Irish
Research Council for Science, Engineering and Technology and Science Foundation
Ireland Grant 07/IN.1/I901.
The authors are with the Hamilton Institute, National University of Ireland
Maynooth, Maynooth, Ireland (e-mail: tianji.li[at]nuim.ie; doug.leith[at]nuim.ie;
david.malone[at]nuim.ie).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TNET.2010.2089992
the Appendix for further details of the configuration used.
We recorded round-trip times (RTTs) before and after one
wireless station started to download a 37-MB file from a Web
site. Before starting the download, we pinged the access point
(AP) from a laptop five times, each time sending 100 ping
packets. The RTTs reported by the ping program were between
2.6–3.2 ms. However, after starting the download and allowing
it to continue for a while (to let the congestion control algorithm
of TCP probe for the available bandwidth), the RTTs to the
AP increased by three orders of magnitude to 2900–3400 ms.
During the test, normal services such as Web browsing experienced
obvious pauses/lags on wireless stations using the
network. Closer inspection revealed that the buffer occupancy
at the AP exceeded 200 packets most of the time and reached
250 packets from time to time during the test. Note that the
increase in measured RTT could be almost entirely attributed
to the resulting queuing delay at the AP and indicates that
a more sophisticated approach to buffer sizing is required.
Indeed, using the algorithm proposed in this paper, the
RTTs observed when repeating the same experiment fall to
only 90–130 ms. This reduction in delay does not come at the
cost of reduced throughput; i.e., the measured throughput with
the algorithm and the default buffers is similar.
In this paper, we consider the sizing of buffers in
802.11/802.11e-based [1], [2] WLANs.We focus on single-hop
WLANs since these are rapidly becoming ubiquitous as the last
hop on home and office networks as well as in so-called “hot
spots” in airports and hotels, but note that the proposed schemes
can be easily applied in multihop wireless networks. Our main
focus in this paper is on TCP traffic since this continues to
constitute the bulk of traffic in modern networks (80%–90%
[35] of current Internet traffic and also WLAN traffic [28]),
although we extend consideration to UDP traffic at various
points during the discussion and also during our experimental
tests.
Compared to sizing buffers in wired routers, a number of
fundamental new issues arise when considering 802.11-based
networks. First, unlike wired networks, wireless transmissions
are inherently broadcast in nature, which leads to the packet
service times at different stations in a WLAN being strongly
coupled. For example, the basic 802.11 DCF ensures that the
wireless stations in a WLAN win a roughly equal number of
transmission opportunities [19], hence the mean packet service
time at a station is an order of magnitude longer when 10 other
stations are active than when only a single station is active.
Consequently, the buffering requirements at each station would
also differ, depending on the number of other active stations
in the WLAN. In addition to variations in the mean service
time, the distribution of packet service times is also strongly dependent
on the WLAN offered load. This directly affects the
burstiness of transmissions and hence buffering requirements
(see Section III for details). Second, wireless stations dynamically
adjust the physical transmission rate/modulation used in
1063-6692/$26.00 © 2010 IEEE
LI et al.: BUFFER SIZING FOR 802.11-BASED NETWORKS 157
order to regulate noncongestive channel losses. This rate adaptation,
whereby the transmit rate may change by a factor of 50
or more (e.g. from 1 to 54 Mb/s in 802.11a/g), may induce large
and rapid variations in required buffer sizes. Third, the ongoing
802.11n standards process proposes to improve throughput efficiency
by the use of large frames formed by aggregation of
multiple packets [3], [18]. This acts to couple throughput efficiency
and buffer sizing in a new way since the latter directly
affects the availability of sufficient packets for aggregation into
large frames.
It follows from these observations that, among other things,
there does not exist a fixed buffer size that can be used for sizing
buffers in WLANs. This leads naturally to consideration of dynamic
buffer-sizing strategies that adapt to changing conditions.
In this paper, we demonstrate the major performance costs associated
with the use of fixed buffer sizes in 802.11 WLANs
(Section III) and present two novel dynamic buffer-sizing algorithms
(Sections IV and V) that achieve significant performance
gains. The stability of the feedback loop induced by
the adaptation is analyzed, including when cascaded with the
feedback loop created by TCP congestion control action. The
proposed dynamic buffer-sizing algorithms are computationally
cheap and suited to implementation on standard hardware. Indeed,
we have implemented the algorithms in both the NS-2
simulator and the Linux MadWifi driver [4]. In this paper, in
addition to extensive simulation results, we also present experimental
measurements demonstrating the utility of the proposed
algorithms in a test bed located in office environment and
with realistic traffic. The latter includes a mix of TCP and UDP
traffic, a mix of uploads and downloads, and a mix of connection
sizes.
The remainder of the paper is organized as follows. Section II
introduces the background of this work. In Section III, simulation
results with fixed-size buffers are reported to further
motivate this work. The proposed algorithms are then detailed
in Sections IV and V. Experiment details are presented in
Section VI. After introducing related work in Section VII, we
summarize our conclusions in Section VIII.
II. PRELIMINARIES
A. IEEE 802.11 DCF
IEEE 802.11a/b/g WLANs all share a common MAC algorithm
called the Distributed Coordinated Function (DCF),
which is a CSMA/CA-based algorithm. On detecting the
wireless medium to be idle for a period , each wireless
station initializes a backoff counter to a random number selected
uniformly from the interval [0, CW-1] where CW is the
contention window. Time is slotted and the backoff counter is
decremented each slot that the medium is idle. An important
feature is that the countdown halts when the medium is detected
busy and only resumes after the medium is idle again for a
period . On the counter reaching zero, a station transmits
a packet. If a collision occurs (two or more stations transmit
simultaneously), CW is doubled and the process repeated. On
a successful transmission, CW is reset to the value and
a new countdown starts.
B. IEEE 802.11e EDCA
The 802.11e standard extends the DCF algorithm (yielding
the EDCA) by allowing the adjustment of MAC parameters that
Fig. 1. WLAN topology used in simulations. Wired backhaul link bandwidth
100 Mb/s. MAC parameters of the WLAN are listed in Table I.
TABLE I
MAC/PHY PARAMETERS USED IN SIMULATIONS, CORRESPONDING TO 802.11g
were previously fixed. In particular, the values of (called
in 802.11e) and may be set on a per-class basis
for each station. While the full 802.11e standard is not implemented
in current commodity hardware, the EDCA extensions
have been widely implemented for some years.
C. Unfairness Among TCP Flow
Consider a WLAN consisting of client stations each carrying
one TCP upload flow. The TCP ACKs are transmitted by
the wireless AP. In this case, TCP ACK packets can be easily
queued/dropped due to the fact that the basic 802.11 DCF ensures
that stations win a roughly equal number of transmission
opportunities. Namely, while the data packets for the flows
have an aggregate share of the transmission opportunities,
the TCPACKs for the flows have only a share.
Issues of this sort are known to lead to significant unfairness
among TCP flows, but can be readily resolved using 802.11e
functionality by treating TCP ACKs as a separate traffic class
that is assigned higher priority [15]. With regard to throughput
efficiency, the algorithms in this paper perform similarly when
the DCF is used and when TCP ACKs are prioritized using the
EDCA as in [15]. Per-flow behavior does, of course, differ due
to the inherent unfairness in the DCF, and we therefore mainly
present results using the EDCA to avoid flow-level unfairness.
D. Simulation Topology
In Sections III, IV, and V-G, we use the simulation topology
shown in Fig. 1, where the AP acts as a wireless router between
the WLAN and the Internet. Upload flows originate from stations
in the WLAN on the left and are destined to wired host(s)
in the wired network on the right. Download flows are from
the wired host(s) to stations in the WLAN. We ignore differences
in wired bandwidth and delay from the AP to the wired
hosts, which can cause TCP unfairness issues on the wired side
(an orthogonal issue) by using the same wired-part RTT for all
flows. Unless otherwise stated, we use the IEEE 802.11g PHY
parameters shown in Table I and the wired backhaul link bandwidth
is 100 Mb/s with RTT 200 ms. For TCP traffic, the widely
deployed TCP Reno with SACK extension is used. The advertised
window size is set to be 4096 packets (each has a payload
158 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 19, NO. 1, FEBRUARY 2011
Fig. 2. Measured distribution of per-packet MAC service time. Solid vertical
lines mark the mean values of distributions. Physical layer data/basic rates are
11/1 Mb/s. (a) 2 stations. (b) 12 stations.
of 1000 bytes), which is the default size of current Linux kernels.
The maximum value of the TCP smoothed RTT measurements
(sRTT) is used as the measure of the delay experienced
by a flow.
III. MOTIVATION AND OBJECTIVES
Wireless communication in 802.11 networks is time-varying
in nature; i.e., the mean service time and the distribution of
service time at a wireless station vary in time. The variations are
primarily due to: 1) changes in the number of active wireless
stations and their load (i.e., offered load on the WLAN); and
2) changes in the physical transmit rate used (i.e., in response
to changing radio channel conditions). In the latter case, it
is straightforward to see that the service time can be easily
increased/decreased using low/high physical layer rates. To see
the impact of offered load on the service time at a station, Fig. 2
plots the experimentally measured distribution of the MAC
layer service time when there are two and 12 stations active.
It can be seen that the mean service time changes by over an
order of magnitude as the number of stations varies. Observe
also from these measured distributions that there are significant
fluctuations in the service time for a given fixed load. This is a
direct consequence of the stochastic nature of the CSMA/CA
contention mechanism used by the 802.11/802.11e MAC.
This time-varying nature directly affects buffering requirements.
Fig. 3 plots link utilization1 and max sRTT (propagation
plus smoothed queuing delay, with smoothing via the standard
TCP RTT filter) versus buffer size for a range of WLAN offered
loads and physical transmit rates.We can make a number of observations.
First, it can be seen that as the physical-layer transmit rate
is varied from 1 to 216 Mb/s, the minimum buffer size to ensure
at least 90% throughput efficiency varies from about 20 to
about 800 packets. No compromise buffer size exists that ensures
both high efficiency and low delay across this range of
transmit rates. For example, a buffer size of 80 packets leads to
RTTs exceeding 500 ms (even when only a single station is active,
and so there are no competing wireless stations) at 1 Mb/s
and throughput efficiency below 50% at 216 Mb/s. Note that
the transmit rates in currently available draft 802.11n equipment
already exceed 216 Mb/s (e.g., 300 Mb/s is supported by
current Atheros chipsets) and the trend is toward still higher
transmit rates. Even across the restricted range of transmit rates
1Here, the AP throughput percentage is the ratio between the actual
throughput achieved using the buffer sizes show on the x-axis and the maximum
throughput over the range of buffer sizes shown on the x-axis.
1–54 Mb/s supported by 802.11a/b/g, it can be seen that a buffer
size of 50 packets is required to ensure throughput efficiency
above 80%, yet this buffer size induces delays exceeding 1000
and 3000 ms at transmit rates of 11 and 1 Mb/s, respectively.
Second, delay is strongly dependent on the traffic load and
the physical rates. For example, as the number of competing
stations (marked as “uploads” in the figure) is varied from 0 to
10, for a buffer size of 20 packets and physical transmit rate
of 1 Mb/s, the delay varies from 300 to over 2000 ms. This
reflects that the 802.11 MAC allocates available transmission
opportunities equally on average among the wireless stations,
and so the mean service time (and thus delay) increases with the
number of stations. In contrast, at 216 Mb/s, the delay remains
below 500 ms for buffer sizes up to 1600 packets.
Our key conclusion from these observations is that there
exists no fixed buffer size capable of ensuring both high
throughput efficiency and reasonable delay across the range
of physical rates and offered loads experienced by modern
WLANs. Any fixed choice of buffer size necessarily carries
the cost of significantly reduced throughput efficiency and/or
excessive queuing delays.
This, therefore, naturally leads to the consideration of adaptive
approaches to buffer sizing, which dynamically adjust the
buffer size in response to changing network conditions to ensure
both high utilization of the wireless link while avoiding unnecessarily
long queuing delays.