Improving IEEE 802.11 WLAN: QoS and Throughput Perspective
description
Transcript of Improving IEEE 802.11 WLAN: QoS and Throughput Perspective
Improving IEEE 802.11 WLAN: QoS and Throughput Perspective
Sunghyun Choi, Ph.D.Assistant ProfessorSchool of Electrical EngineeringSeoul National UniversityE-mail: [email protected]: http://ee.snu.ac.kr/~schoi
2
Introduction to My Group in SNU
Multimedia & Wireless Networking Lab. (MWNL) Within School of Electrical Engineering,
Seoul National University
Started September 2003 One of the youngest groups in SoEE, SNU
1 (+2) Ph.D. & 3 masters students
3
Introduction to My Group in SNU (Cont’d) Working on WLAN MAC and around
Resource management – power, rate, … QoS & mobility TCP/UDP over WLAN
4G wireless network Cross-layer design (Sensor networks)
4
Contents
Introduction
QoS provisioning
Throughput enhancement
Conclusion
5
Introduction to IEEE 802.11 WLAN Wireless Ethernet with comparable speed
Supports up to 11 and/or 54 Mbps within >100 m range
Enable (indoor) wireless and mobile high-speed networking Runs at unlicensed bands at 2.4GHz and 5GHz Connectionless MAC and multiple PHYs
6
Limitations of Current 802.11
Lack of QoS support Best-effort service with contention-based MAC
Low throughput due to large overhead < 5 Mbps throughput at 11 Mbps 802.11b link
My group is currently working on improving both aspects Will show only preliminary results here
7
QoS Improvement
8
Emerging IEEE 802.11e MAC
New draft standard for QoS provisioning Expected to be finalized by early next year
Defining a new MAC backward compatible with the legacy MAC Legacy 802.11 MAC – DCF (+ PCF) 802.11e MAC – HCF with two access
mechanisms Controlled channel access Contention-based channel access (EDCA)
9
802.11 Distributed Coordination Function (DCF) Carrier Sense Multiple Access with Collision
Avoidance (CSMA/CA)
BusyMedium
SIFS
PIFS
DIFS
BackoffWindow
Slot Time
Defer Access Select Slot and decrement backoffas long as medium stays idle
DIFS
Contention WindowImmediate access whenmedium is idle >= DIFS
Next Frame
10
802.11e Access Category (AC) Access category (AC)
as a virtual DCF 4 ACs implemented
within a QSTA to support 8 priorities
Multiple ACs contend independently
The winning AC transmits a frame
AC0 AC1 AC2 AC3
Virtual Collision Handler
Backo
ff A
IFS[0]
BO
[0]
Backo
ff A
IFS[1]
BO
[1]
Backo
ff A
IFS[2]
BO
[2]
Backo
ff A
IFS[3]
BO
[3]
Transmission Attempt
11
Differentiated Channel Access of 802.11e EDCA Each AC contentds with
AIFS[AC] (instead of DIFS) and CWmin[AC] / CWmax[AC] (instead of CWmin / CWmax)
BusyMedium
SIFS
PIFS
AIFS[AC]
BackoffWindow
SlotTime
Defer Access Select Slot and decrement backoffas long as medium stays idle
AIFS[AC]+SlotTime
Contention Windowfrom [1,1+CWmin[AC]]
Immediate access whenmedium is idle >=AIFS[AC]+SlotTime
Next Frame
12
Simulation Results - DCF vs. EDCA Delay comparison
2 video (1.5 Mbps CBR), 4 voice (36.8 kbps CBR), 4 data (1 Mbps Poisson)
13
Our Software-Based Approach for RT Traffic Support
IEEE 802.11e is not available yet Even if it becomes available, many
existing legacy 802.11 APs will be there Especially, for WISP with many deployed
APs, replacing existing APs costs a lot of money
Software (or firmware) upgrade-based approach is very desirable at least in the short term
14
System Architecture
Device Driver
TCP/UDP
IP
PHY
MAC
RT+NRT
frame processing
TCP/UDP
IP
PHY
MAC
RT+NRT
RT NRT
frame processing
(a) Original Host AP driver
(b) Modified Host AP driver
15
Measurement Configuration
Linux + HostAP driver for Intersil chipsets one RTP (1.448 Mbps CBR) + one FTP
Console
Server
Host AP Clientswitch
Implement dual queue
16
One-Way Delay of RTP Traffic
Original
Modified
17
Percentage Gain in Performance Parameters
Test 1
ComparisonPercentage
gainOriginal
Two Queue
Throughput
TCP 3.851 3.703 -3.84%
RTP 1.448 1.448 0.00%
Jitter of RTP
Avg. 2.9 2.6 -10.34%
Max.
4.0 3.0 -25.00%
min. 2.0 2.0 0.00%
One-way delay of
RTP
Avg. 30.7 20.2 -34.20%
Max.
32.0 23.0 -28.13%
min. 32.0 18.0 -40.00%
Max delay variation of RTP
27.0 18.0 -33.33%
Percentage Gain in Performance Parameters
TCP throughput
RTP troughput
Jitter Avg.
Jitter Max.
Jitter Min
One-way delay Avg.
One-way delay Max.
One-way delay Min.
Max delay variation
-70%
-60%
-50%
-40%
-30%
-20%
-10%
0%
10%
18
Limitations and Future Work
Limitations of the current approach Running on top of legacy MAC with a single
FIFO queue AP cannot prevent/control contention from
stations Downlink RT transmission could be severely
delayed due to the uplink contentions
How to handle this situation is an on-going effort
19
Throughput Improvement
20
IEEE 802.11n Initiative
A new standardization effort to achieve over 100 Mbps throughput over WLAN
Via both PHY and MAC enhancement
We are considering the MAC improvement for throughput enhancement
21
Frame Size Affects Throughput 802.11 MAC/PHY have big fixed overheads
MAC header, IFSs, ACK, and Backoff PLCP preamble & header
Busy Channel
DIFS34
usec PPDU ACK
PLCPPreamble
PLCP Header
Payload FCS
16 usec >= 4 usec 24 octets Variable 4 octets
SIFS (16 usec)
Backoff - 9 usec x [0,CW] >= 24 usec
MAC Header
22
0
5
10
15
20
25
30
35
40
Datagram Size (octets)
Thro
ughp
ut (
Mbp
s)Theoretical Throughput for 54 Mbps
Preferred
Operation Range
23
Packet Size Statistics
This statistics is from the measurement taken in the 802.11 standard meeting room in the morning of July 22nd 2003
24
Frame Aggregation
Aggregation of multiple frames in order to reduce the fixed overheads relatively!
Multiple frames are aggregated above the MAC SAP The aggregated frame is transmitted via a data frame
25
Frame Formats
octets: 2 2 6 6 6 2 variable 4
octets: 3 5 variable IP
datagramSNAP
HeaderLLC
Header
Frame Control
Duration / ID Addr 1 Addr 2 Addr 3
Seq. Control Data FCS
802.2 LLC
802.11 MAC
octets: 2 2 6 6 6 2 variable 4
octets: 1 1 1 1 2 2 2 variable variable
Frame Control
Duration / ID
Addr 1
Addr 2
Addr 3
Seq. Control Data FCS
802.2 LLC with aggregation
802.11 MAC
Control(0x03)
Count(n)
Size 1 ... Size nEtherframe
1Etherframe
n...
octets: 6 6 2 variableSource
Addr.Type IP datagram
DestinationAddr.
SSAP(0xdd)
DSAP(0xdd)
Reserved
Original
With aggregation
26
Theoretical Throughput w/ Aggregation (w/o channel error)
0
5
10
15
20
25
30
35
40
Datagram Size (octets)
Thro
ughp
ut (
Mbp
s)
11a w/o aggregation 11a w/ aggregation # of aggregated datagrams
27
Theoretical Throughput w/ Aggregation (w/ channel error)
28
Performance Measurement
Implement frame aggregation in real platform Linux & Intersil-based platform (.11b) Measure the throughput performance of UDP
and TCP traffic Note: Frame aggregation is only applied when
there are multiple frames in the queue
Traffic generator AP STA
29
Measurement Results - UDP
Packet aggregation, RTP, 10Mbps
0
1
2
3
4
5
6
7
8
0 200 400 600 800 1000 1200 1400 1600Packet Size (bytes, application-level)
Thr
ough
put (
Mbp
s)
original
aggregation
Theoretical
Throughput performance of packet aggregation with fixed rate UDP
30
Measurement Results - TCP Throughput performance of packet aggregation with
TCP
Packet Aggregation, TCP
0
1
2
3
4
5
6
7
0 200 400 600 800 1000 1200 1400 1600Packet Size (bytes, application-level)
Thr
ough
put (
Mbp
s)
original
aggregation
Theoretical
31
Summary and Future Work
Shown that frame aggregation is a good way to improve 802.11 MAC throughput Via both analysis and measurements
Frame aggregation can be done above the MAC SAP very easily
Needs further measurements/simulations for more realistic scenarios
32
Concluding Remarks
IEEE 802.11 WLAN is becoming real popular these days
There is still a big room to improve the current 802.11 systems
Important to consider how any improved system co-exists with legacy systems