Overload Prediction Based on Delay in Wireless OFDMA Systems

E. O. Lucena, F. R. M. Lima, W. C. Freitas Jr and F. R. P. Cavalcanti

Federal University of Ceará - UFC, Wireless Telecommunications Research Group – GTEL CP 6005, Campus do Pici, 60455-760, Fortaleza-CE, Brazil

IEEE Globecom 2010

Speaker ： Tsung-Yin Lee

2

Outline

Introduction Adaptive Congestion Control Framework

Admission Control Scheduling Load Control

Overload Prediction Based on Delay Simulation and Conclusion

3

Introduction

Paper deal with Congestion Control (CC) strategies to protect the QoS of Real Time services in OFDMA-based system (LTE, WiMAX)

The work [3] proposed a QoS-driven adaptive CC framework Admission Control Scheduling Load Control

[3] E. B. Rodrigues and F. R. P. Cavalcanti, “QoS-Driven Adaptive Congestion Control for Voice over IP in Multiservice Wireless Cellular Networks,” IEEE Communications Magazine, vol. 46, no. 1, pp. 100–107, January 2008.

4

Contribution

the generalization of the CC framework proposed in [3] to work with multiple subcarriers to be considered for networks employing OFDMA in the downlink (AdaptiveCC)

a new feature based on delay to be added to the generalized framework to predicts an overload situation (Delay-based Prediction)

Simplified system architecture and the CC framework

5

6

Session Admission Control (SAC)

Session Admission Control (SAC) scheme is employed to guarantee the quality of a RT service in a mix with other services The SAC algorithm considers delay as the

resource to be shared among flows in the system the packet delays of the RT traffic are regularly

measured and filtered by an Attack-Decay (AD) Filter (discard or transmit packet)

AC procedure

There are two admission thresholds depending on the service type : for the RT service and for other low

priority services the SAC algorithm will check if the filtered

delay calculated by the AD filter (DRT) If the new flow is rejected; otherwise

the flow is admitted (same as NRT service)

7

thRTD th

OtherD

thRTRT DD

SAC Priority Margin

Adaptive CC framework is to adapt them according to the congestion status of an RT service the SAC priority margin, α. Paper defines this priority

margin (in decibel) as

α[k] is the SAC priority margin at Transmission Time Interval (TTI) k

8

Ex: If VoIP service is and α = –0.5 dB, the SAC admission threshold for the Web service is given by msDD th

RTthWeb 69.13310 10

msD thRT 150

9

Scheduling (1/2)

Weighted Proportional Fair (WPF), the scheduled flow is the one with highest priority. The priority of a flow j at TTI k is given by

wj[k] represents a service-dependent weight

rj[k] is the supported data rate

tj[k] is the filtered data rate according to the channel state

Scheduling (2/2)

In an OFDMA-based system, the prioritization in Weighted Multicarrier Proportional Fair (WMPF) is given by

The flow selection consists in assigning the pair flow-subcarrier corresponding to the largest entry in the priority matrix

10

subcarrier

Adaptive Scheduling

if the flow is from an RT session, wj[k] is set to WRT ,and on the other hand, if the flow is from another service the weight is equal to Wother

the Adaptive CC framework adapts Wother to control the congestion in the RT service (In [2], WRT ,and Wother are fixed)

11

[2] A. R. Braga, E. B. Rodrigues, and F. R. P. Cavalcanti, “Packet Scheduling for VoIP over HSDPA in Mixed Traffic Scenarios,” in Personal, Indoor and Mobile Radio Communications, 2006 IEEE 17th International Symposium on, Helsinki, September 2006, pp. 1–5.

WMPF Priority Margin

Paper define β[k] priority margin (in dB) as

WRT and WOther[k] represent service-dependent weights of the RT flow and other service, respectively, and β[k] is the value of β at TTI k

12

Ex: If VoIP service is and β = –0.5 dB, the WPF priority weight for Web service is given by 89.010 10

VoIPWeb WW1VoIPW

13

Load Control (LC)

the priority margins α and β can control the prioritization of the RT service over other services (WRT and are fixed)

the LC algorithm is to adapt these priority margins according to the QoS of the ongoing sessions of the high priority service

thRTD

Adaptation Priority Margin

The adaptation of the priority margin α[k] is given as follows: (same as β[k])

The FER considers a ratio of number of lost frames (or packets) and the total number of generated packets

αmin, αmax are the minimum and maximum values in dB

The fixed parameters σα control the adaptation speed14

Adaptation of the WPF priority margin over time depending on the filtered VoIP FER

15

Congestion control framework operation

16

17

Overload Prediction Based on Delay

An early detection of overload situations based on the packet delay of RT flows in a framework called Delay-based Prediction Motivation : as the reaction of the Adaptive CC

takes place when the overload already exist, the system will get back to normal load conditions only after a certain period

18

Delay Prediction Variable Y

The increasing delay information (Y) can be added to the priority margins α and β

The variable Y works adding a value to the equation and it can be modified and yields

19

Behavior of the Delay Prediction Variable Y

M is a constant responsible for the slope of the exponential curve

Ymin and Ymax are fixed parameters that indicate the minimum and the maximum values of Y

Y increases only when the monitored delay is higher than a threshold D0

20

Performance Evaluation

In this section, paper present a performance evaluation of the Delay-based Prediction framework compared to a framework without overload prediction called Adaptive CC and to a reference framework called Non Adaptive CC

Main Parameters of the Simulation Tool

21

Performance Metrics

For a VoIP flow, the satisfaction is reached when it was not blocked by the AC functionality and its experienced FER is equal to or lower than 1%

For the WWW service, the flow is satisfied when it was not blocked and its average throughput assumes at least the rate requirement value of 128 kbps

22

1.375 Users/s for mix 25% of VoIP and 75% of WWW flows

23

Satisfaction for three different frameworks

24

25

Conclusion

In this work paper presented two contributions to protect the QoS of RT services in a mixed traffic scenario the generalization of the CC framework

proposed (Scheduling, Admission Control, Load Control)

The Delay-based Prediction framework besides guaranteeing the QoS fulfillment of a RT service also prevent high peaks of FERs