Quality of Service Schemes Quality of Service Schemes for IEEE 802.11 Wireless for IEEE 802.11 Wireless

LANs-An EvaluationLANs-An Evaluation主講人主講人 : : 黃政偉黃政偉

OutlineOutline IntroductionIntroduction Overview of evaluated schemesOverview of evaluated schemes SimulationSimulation ResultResult DiscussionDiscussion ConclusionConclusion

IntroductionIntroduction Wireless networks are superior to wired networks with regard Wireless networks are superior to wired networks with regard

to aspects such as to aspects such as ease of installation and flexibilityease of installation and flexibility.. Given Given the coverage and low pricethe coverage and low price, it is likely that demands for , it is likely that demands for

the ability to run real-time applications such as voice over IP the ability to run real-time applications such as voice over IP over these networks will increase.over these networks will increase.

Inherent problems: low medium utilization, risk of collisions Inherent problems: low medium utilization, risk of collisions and problem of providing differentiation between different and problem of providing differentiation between different types of traffic.types of traffic.

Overview of evaluated schemesOverview of evaluated schemes Point Coordinator Function (PCF)Point Coordinator Function (PCF)

Overview of evaluated schemes Overview of evaluated schemes Enhanced DCF(EDCF)Enhanced DCF(EDCF)

The EDCF mechanism allows traffic to be classified iThe EDCF mechanism allows traffic to be classified into nto 88 different traffic classes, different traffic classes,

MethodMethod: the : the minimum contention window (CWminimum contention window (CWminmin)) an and d the interframe spacethe interframe space

Higher priorityHigher priority : : smaller CWsmaller CWmin min ; ; lowlower priorityer priority : : larger CWlarger CWminmin

Different traffic classesDifferent traffic classes: : Arbitration interframe space Arbitration interframe space (AIFS = DIFS + number of time slots)(AIFS = DIFS + number of time slots)

Overview of evaluated schemes Overview of evaluated schemes Enhanced DCF(EDCF)Enhanced DCF(EDCF)

Overview of evaluated schemes Overview of evaluated schemes Distributed Fair Scheduling (DFS)(1/2)Distributed Fair Scheduling (DFS)(1/2)

Fair means that each flow gets bandwidth proportional to sFair means that each flow gets bandwidth proportional to some weight that has been assigned to it.ome weight that has been assigned to it.

Before transmitting a frame, the backoff process is always Before transmitting a frame, the backoff process is always initiatedinitiated

Where sizeWhere sizepacketpacket is the size of the packet to send is the size of the packet to send

packetsize

B ctorScaling_fa

magnitude. suitable of valuestointervals backoff thescale toused isctor Scaling_Fa

1mean with variablerandom a is ,station theof weight theis

Overview of evaluated schemes Overview of evaluated schemes DistriDistributed Fair Scheduling (DFS)(2/2)buted Fair Scheduling (DFS)(2/2)

Fairness is achieved by using the size of the packet to Fairness is achieved by using the size of the packet to be sent in the calculation of the backoff interval.be sent in the calculation of the backoff interval.

If a collision occurs, a new backoff interval is calculatIf a collision occurs, a new backoff interval is calculated using the 802.11 standard. (CWed using the 802.11 standard. (CWminmin=3)=3)

Overview of evaluated schemes Overview of evaluated schemes BlackBlackburst (BB)(1/2)burst (BB)(1/2)

Three interframe spacing: Three interframe spacing: ((ττ: max propagation delay): max propagation delay) ttshort short + 2+ 2ττ< t< tmed med ; t; tmed med + 2+ 2ττ< t< tlonglong ; ; ttshort short < < ttmedmed < t < tlonglong

Black burst duration:Black burst duration: Duration = tDuration = tbslotbslot * [d / t * [d / tunitunit]; t]; tbslotbslot is a length of a black slot . is a length of a black slot .

ttunitunit : a system parameter defined shortly. : a system parameter defined shortly.

d : to access the medium, STA has been waiting for d sec.d : to access the medium, STA has been waiting for d sec. The scheduled access interval : tThe scheduled access interval : tsch sch

Overview of evaluated schemes Overview of evaluated schemes BlackBlackburst (BB)(2/2)burst (BB)(2/2)

Simulation Simulation Scenarios(1/2)Scenarios(1/2)

Network simulator : ns-2Network simulator : ns-2 WLAN bandwidth : 2Mbit/sWLAN bandwidth : 2Mbit/s In infrastructure mode, the mobile nodes always communicate In infrastructure mode, the mobile nodes always communicate

directly with the AP.directly with the AP. There is no mobility in the systemThere is no mobility in the system

Packet size Packet size (m, (m, σσ22))

Inter-packet Inter-packet intervalinterval

Bit rateBit rate

High priorityHigh priority 300bytes300bytes40 bytes40 bytes

25-40 ms25-40 ms 60-96 kbit/s60-96 kbit/s

Low priorityLow priority 800 bytes800 bytes150 bytes150 bytes

50ms50ms 128 kbit/s128 kbit/s

Simulation Simulation Scenarios(2/2)Scenarios(2/2)

table1table1

TU = 10TU = 102424μμss

SimulationsSimulationsMetrics(1/2)Metrics(1/2)

Throughput (Normalization)Throughput (Normalization) Percentage of the offered data that is actually delivered to Percentage of the offered data that is actually delivered to

the destination.the destination. Medium utilizationMedium utilization

How large percentage of time that is used for successful How large percentage of time that is used for successful transmission of data framestransmission of data frames..

Collision rateCollision rate The average number of collisions that occur per second.The average number of collisions that occur per second.

SimulationsSimulationsMetrics(2/2)Metrics(2/2)

Access delayAccess delay as the time the Head-of-Line data packet spends at the as the time the Head-of-Line data packet spends at the

MAC layer before being successfully transmitted out on MAC layer before being successfully transmitted out on the wireless medium.the wireless medium.

Cumulative delay distributionCumulative delay distribution We present the cumulative distribution of the access delays We present the cumulative distribution of the access delays

for high priority traffic to find out the percentage of the for high priority traffic to find out the percentage of the packets that are below certain delay bounds.packets that are below certain delay bounds.

ResultResultDetermining PCF superframe size(1/3)Determining PCF superframe size(1/3)

Size for Size for high priority traffichigh priority traffic Short control frame polled delay Short control frame polled delay

throughputthroughput long control frame polled delay long control frame polled delay

throughputthroughput The best performance for high priority traffic would be achThe best performance for high priority traffic would be ach

ieved by having a superframe size similar to ieved by having a superframe size similar to the interval bethe interval between the frames generated by the nodes.tween the frames generated by the nodes.

ResultResultDetermining PCF superframe size(2/3)Determining PCF superframe size(2/3)

ResultResultDetermining PCF superframe size(3/3)Determining PCF superframe size(3/3)

Superframe of 20 time units for PCF in the comparison of the different QoS shcemes

ResultResult1.Throughput(1/2)1.Throughput(1/2)

The objective of DFS is to provide fair differentiation.

DFS always allocates a share of the bandwidth for low priority traffic and avoids starvation.

ResultResult1.Throughput(2/2)1.Throughput(2/2)

The difference in performance between BB, PCF and EDCF is quite small.

Both EDCF and BB starve low priority traffic rather fast, and PCF only gives a very small share of the bandwidth to low priority traffic.

ResultResult2. Medium utilization2. Medium utilization

ResultResult3. Collision Rate3. Collision Rate

ResultResultOverhead by BB and PCFOverhead by BB and PCF

ResultResult4. Access delay(1/2)4. Access delay(1/2)

ResultResult4. Access delay(2/2)4. Access delay(2/2)

ResultResult5. Cumulative delay distribution(1/2)5. Cumulative delay distribution(1/2)

ResultResult5. Cumulative delay distribution(2/2)5. Cumulative delay distribution(2/2)

DiscussionDiscussion Admission control : PCF easily implements.Admission control : PCF easily implements. Starving low priority traffic: BB and EDCF.Starving low priority traffic: BB and EDCF. If Blackburst could not be used, EDCF could be a suitIf Blackburst could not be used, EDCF could be a suit

able alternative. able alternative. Comparing EDCF and DFS at high loads. In Fig. 5 anComparing EDCF and DFS at high loads. In Fig. 5 an

d 8, EDCF has both higher throughput and lower averd 8, EDCF has both higher throughput and lower average delay than DFS for high priority traffic.age delay than DFS for high priority traffic.

Different settings used to create the different scenarioDifferent settings used to create the different scenarios of course affect the final result.s of course affect the final result.

ConclusionsConclusions When using PCF, it is important to select a proper size of the sWhen using PCF, it is important to select a proper size of the s

uperframe. uperframe. The superframe should be approximately as long as the intervaThe superframe should be approximately as long as the interva

l between packets generated by a high priority station.l between packets generated by a high priority station. Blackburst gives the best performance with regard to throughpBlackburst gives the best performance with regard to throughp

ut and access delay.ut and access delay. Both Blackburst and EDCF starve low priority traffic at high lBoth Blackburst and EDCF starve low priority traffic at high l

oads of high priority traffic, which in many cases is not desiraoads of high priority traffic, which in many cases is not desirable. In the case, DFS can do a better job.ble. In the case, DFS can do a better job.