37
Providing QoS in Ad Hoc Networks with Distributed Resource Reservation IEEE802.11e and extensions Ulf Körner and Ali Hamidian

Providing QoS in Ad Hoc Networks with Distributed Resource Reservation IEEE802.11e and extensions Ulf Körner and Ali Hamidian

Embed Size (px)

Citation preview

Providing QoS in Ad Hoc Networks with Distributed Resource

Reservation

IEEE802.11e and extensions

Ulf Körner and Ali Hamidian

The Goal

• To provide QoS guarantees to WLANs operating in ad hoc mode– by allowing stations to reserve resources

(medium time)– by distributing the existing admission control

and scheduling algorithms

• Example of application area: gaming

No QoS in IEEE 802.11

• Today’s WLANs do not offer any QoS– usually not a big

problem if you just surf the Internet

– bad voice/video quality if you use e.g. Skype or MSN messenger

802.11 MAC & its QoS Limitations

• 802.11 has two medium access methods:– distributed coordination function (DCF)

• All data flows have the same priority

– point coordination function (PCF)• Not possible for stations to send QoS requirements to the AP• Unknown transmission time of the polled stations

• 802.11e introduces:– hybrid coordination function (HCF)

• enhanced distributed channel access (EDCA)• HCF controlled channel access (HCCA)

Hybrid Coordination Function (HCF)

• transmission opportunity (TXOP): A bounded time interval during which a station may transmit multiple frames– Solves the PCF problem with unknown transmission times

• traffic specification (TSPEC): Contains information about the QoS expectation of a traffic stream (frame size, service interval, data rate, burst size, delay bound, etc.)– Solves the PCF problem with the inability to send QoS

needs

Enhanced Distributed Channel Access (EDCA)

• Contention-based• “Enhanced DCF”• access category (AC):

Each station has four ACs (”transmission queues”). Each AC contends for TXOPs independently of the other ACs

• Service differentiation is realized by varying – Different parameters

AIF

SN

[1]C

Wm

in[1]C

Wm

ax[1]T

XO

Plim

it[1]

AIF

SN

[2]C

Wm

in[2]C

Wm

ax[2]T

XO

Plim

it[2]

AIF

SN

[4]C

Wm

in[4]C

Wm

ax[4]T

XO

Plim

it[4]

AIF

SN

[3]C

Wm

in[3]C

Wm

ax[3]T

XO

Plim

it[3]

virtual collision handler

Background [1] Best effort [2] Video [3] Voice [4]

mapping to AC

HCF Controlled Channel Access (HCCA)

• Contention-free• “Enhanced PCF”• Medium access controlled by a QoS access

point (QAP)• HCCA allows stations with QoS traffic to reserve

TXOPs using TSPECs

Motivation of our Work: QoS Limitations in 802.11e

• Problem with EDCA– Random medium access & no distributed admission

control => not possible to guarantee QoS

• Problem with HCCA– Centralized infrastructure requirement => HCCA not

useful in ad hoc networks

• We need a solution which is– Deterministic (unlike EDCA)

• Remove the random medium access delays

– Distributed (unlike HCCA)• Remove the need of an access point

EDCA with Resource Reservation (EDCA/RR)

• distributed admission control and scheduling

• possibility to reserve TXOPs for deterministic and contention-free medium access

EDCA/RR Operation

Similar to EDCA as long as LP frames (AC_Background and AC_BestEffort) are sent

EDCA/RR Operation

When a HP frame (AC_Video and AC_Voice) reaches the MAC sublayer, the source checks whether its new stream can be admitted

EDCA/RR Operation

If admission control OK: • schedule the new

stream• broadcast ADDTS

request containing TSPEC

• wait for ADDTS response

ADDTS request

EDCA/RR Operation

Once all ADDTS responses are received by the source, it waits until its first reserved TXOP at service start time & starts transmitting

ADDTS response

EDCA/RR Operation

deterministic and contention-free medium access: the source has now reserved TXOPs every scheduled service interval (SI)

HP data frames

Results

• EDCA/RR implementation in ns-2 based on an enhanced 802.11/802.11e implementation

• EDCA vs. EDCA/RR• Stationary behaviour: How is the average end-

to-end delay of a HP-stream affected when the number of LP streams increases?

Throughput: EDCA

1 LP-stream and 4 HP-streams each started 10 s apart.

ad hoc network

Throughput: EDCA

1 LP-stream and 4 HP-streams each started 10 s apart.

ad hoc network

1 LP stream

Throughput: EDCA

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 1 HP stream

Throughput: EDCA

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 2 HP streams

Throughput: EDCA

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 3 HP streams

Throughput: EDCA

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 4 HP streams

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 1 admitted HP stream

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 2 admitted HP streams

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 3 admitted HP streams

Throughput: EDCA/RR

ad hoc network

1 LP-stream and 4 HP-streams each started 10 s apart.

1 LP stream + 3 admitted HP streams +1 rejected HP stream

Throughput: EDCA vs. EDCA/RR

EDCA EDCA/RR

End

Average End-to-End Delay

- 1 HP source- 150 simulation runs! - simulation time: 200 s

Problem due to Hidden Stations

• The hidden station C doesn’t receive A’s ADDTS request so it can start sending just before A’s TXOP starts! ==> no QoS guarantees!

Solving the Hidden Station Problem

• The TSPEC is included in the ADDTS response so when B sends an ADDTS response to A, C hears that message and learns about A’s reservation

• In addition:

Send RTS_TSPEC and CTS_TSPEC in the beginning of each TXOP

Results - 0 % packet error

nbr of LP-streams

average end-to-end delay (ms)

99 % confidence interval (ms)

EDCA EDCA/RR EDCA EDCA/RR

0 0.69 12.33 (0.69,0.69) (12.13,12.53)

1 6.21 12.22 (6.20,6.22) (12.02,12.42)

2 11.17 12.27 (11.14,11.19) (12.08,12.47)

3 13.93 12.22 (13.90,13.96) (12.01,12.42)

4 17.12 12.38 (17.08,17.16) (12.19,12.57)

5 20.51 12.25 (20.46,20.56) (12.06,12.45)

Results - 5 % packet error

nbr of LP-streams

average end-to-end delay (ms)

99 % confidence interval (ms)

EDCA EDCA/RR EDCA EDCA/RR

0 0.99 12.55 (0.99,0.99) (12.37,12.73)

1 4.68 12.44 (4.68,4.69) (12.27,12.61)

2 5.25 12.54 (5.24,5.25) (12.35,12.73)

3 5.59 12.34 (5.58,5.60) (12.16,12.52)

4 5.92 12.64 (5.91,5.93) (12.45,12.82)

5 6.28 12.53 (6.27,6.29) (12.34,12.72)

Results - 0 % packet error

nbr of LP-streams

jitter (10-6 s2) C2[d]

EDCA EDCA/RR EDCA EDCA/RR

0 0.02 48 0.05 0.32

1 40 48 1.04 0.32

2 180 48 1.45 0.32

3 276 48 1.42 0.32

4 406 49 1.38 0.32

5 577 49 1.37 0.32

Multi-hop Resource Reservation

1) A: if traffic is admitted, send RREQ-ADDTSRequest

2) B: if traffic is admitted, send RREQ-ADDTSRequest

3) C: if traffic is admitted, schedule traffic and send RREP-ADDTSResponse

4) B: schedule traffic and send RREP-ADDTSResponse

5) A: schedule traffic and send data

AODV + EDCA/RR

Summary

• EDCA/RR – is a MAC scheme with distributed admission

control and scheduling– allows stations to reserve TXOPs for

deterministic and contention-free medium access