On QoS Guarantees with Reward Optimization for Servicing Multiple

Priority Class in Wireless Networks

YaoChing Peng

Eunyoung Chang

Agenda

Introduction Related Works The Proposed Approach

System model Elastic threshold based Call Admission Control (CAC) algorithms Heuristic-based search algorithm

Performances Analysis Compares elastic threshold-based algorithm against existing CAC algorithms

Conclusion

Introduction

Call admission control (CAC) For single-class network traffic real-time multimedia service such as video and audio and non-real-time services

such as text and image.

The goal is to service multiple priority classes in wireless networks to maximize the system reward rate with QoS guarantees. Dropping probability of handoff calls

Blocking probability of new calls. Handoff occurs when a mobile user with an ongoing connection leaves the current cell

and enters another cell. An ongoing connection maybe dropped when if there is insufficient bandwidth in the

new cell to support to it, Reducing the handoff call drop probability by rejecting new connection requests could

result in an increase in the new call blocking probability.

:

Related Works

Some CAC algorithms proposed that partition system resources and allocate distinct partitioned resources to serve distinct service class.

Some works focus on the distributed CAC algorithm that runs in each cell and partitions channel resources in the cell into three partitions: one for real-time, one for non-real time, and one for both.

Some proposed a bandwidth reservation and reconfiguration mechanism to facilitate handoff processes for multiple service.

Related Works(Cont.)

All CAC algorithms make acceptance decisions for new and handoff calls to satisfy QoS requirement in order to keep the dropping probability of handoff calls and the blocking probability of new calls lower than pre-specified thresholds.

Without considering “value” issues associated with service classes.

Elastic Threshold-Based Call Admission Control (E-CAC)

New threshold-based CAC algorithm has a goal to maximize the system reward rate with QoS guarantees. “Reward” is referring to “value” brought to the system due to services. It could map to “revenue” from the service provider’s perspective.

Elastic threshold-based CAC developed is capable of satisfying QoS while generating higher rewards compared with existing CACs.

cell

cell

System Model

A wireless cellular network

Base station

Mobile user

Service types Example Reward Priority Class

Real time Video, Audio Large reward High priority Class 1

Non-real time Text, Image Small reward Low priority Class 2

we will consider only two service classes in this paper.

New and Handoff call

Cell

Cell 2

System Model

A handoff occurs when a mobile user with an ongoing connectionleaves the current cell and enters another cell.

New call: mobile user send request to current cell

Reward

Handoff New

Class 1

Class2

1 1 2 2T h n h nR R R R R

1hR

2hR 2

nR

1nR

The total reward generated by each cell per unit time would be the sum of the rewards generated by various priority classes

Blocking Probability Thresholds

1 1h hB Bt1 1n nB Bt

2 2h hB Bt

2 2n nB Bt

The QoS constraints are expressed in terms of blocking probability thresholds.

Pre-specified thresholds

1 1h nBt Bt

Dropped handoff calls dissatisfy users more than blocked new calls do, so the drop probability requirement of handoff calls is likely to be more stringent than the blocking probability requirement of new calls

1hB1nB

Handoff dropping probability of class 1

New call blocking probability of class 1

Threshold-Based Admission Control

Without Elastic Threshold-Based

Previous CAC algorithms make acceptance decisions for new and handoff calls to satisfy QoS requirements to keep the dropping probability of handoff calls and the blocking probability of new calls lower than pre-specified thresholds.

without considering value issues associated with service classes.

maximizing the reward of the system through CAC in the context of multimedia services(High priority).

The probabilities of accepting a new call and a handoff call of service class i

1K number of channels required by a service call of service class 1

C the total number of channels

1hHTh high threshold of handoff call at class 1

2nLTh Low threshold of new call at class 2

Simulation

Accepting rate: P when

n:the number of channels that have been allocated in the system

Place: hold service calls

Transition : models the arrival rate

Transition : models the departure rate

Guard

the number of handoff calls it holds

number of channels required by a service call of service class 1

the number of channels that have been allocated in the system

The probabilities of accepting

SPN

The probabilities of accepting

SPN

The probabilities of accepting

SPN

Errors in the paper

P.4

On the line right below Equation 12, "if Ein and Eih are enabled" should be "if Ein and Eih are disabled".

Also right below Equation 14, "if Ein and Eih are disabled" should be "if Ein and Eih are enabled".

by Dr. Chen

Blocking/Dropping Probabilities: SPN

Expected value of a random variable X

Handoff dropping probability of class i

New call blocking probability of class i

Charge-by-timeThe reward earned per unit time

1 1 2 2T h n h nR R R R R

the reward for a call ofservice class i per unit time.

Threshold problem

High threshold of handoff call at class 1

1nLTh Low threshold of new call at class 1

1hBt1nBt

2hBt

2nBt

1nHTh

1hLTh

1hHTh

High threshold of new call at class 1

High threshold of handoff call at class 22nHTh

2hHTh

High threshold of new call at class 2

Low threshold of handoff call at class 12nLTh Low threshold of new call at class 22hLTh Low threshold of handoff call at class 2

Handoff dropping probability of class 1

Handoff dropping probability of class 2

New call blocking probability of class 1

New call blocking probability of class 2

Heuristic-based Search Algorithm

To determine the optimal threshold combinations that would maximize the reward earned per unit time with satisfying QoS constraints.

The E-CAC algorithm utilizes a greedy search method to determine a legitimate solution which maximizes the reward rate.

A delta (∆) value as input : to determine the set of threshold combinations in the range of current threshold ± Δ. Step 1: Finding a legitimate solution

Method1 : finding a legitimate solution which satisfied the QoS constraints of all service call types

Method 2: determine the minimum number of channels needed, Step 2 : determining a locally optimal solution by applying a greedy search

starting from the legitimate solution found in the first step

Search for Optimal Threshold Values

Method 1 : finding a legitimate solution.

Set all thresholds to the total number of channels (namely C), and check if this combination is legitimate.

Reduce the bandwidth used by lowering the low threshold of low-priority classes until find a legitimate solution or start missing QoS constraints of low-priority classes.

Next, start reducing arrivals of high-priority new calls as well by lowering their low threshold.

If none of these approaches generates a legitimate solution, this method returns no legitimate solution.

Search for Optimal Threshold Values

Method 2 : determine the minimum number of channels needed to satisfy the QoS constraint

This helps eliminate all threshold combinations that do not provide the minimum number of channels in the search.

In subsequent iterations, If the low threshold ≤ the high threshold,

Then increase the low thresholds of low-priority calls Otherwise, increase the high threshold

until the high threshold reaches C or find a legitimate solution.

Similarly, increase the low threshold of high-priority new calls If cannot satisfy the QoS constraints of high-priority handoff calls,

then decrease the high thresholds of low-priority calls and the high threshold of high-priority new calls

At any point, if determine that we have a threshold combination, break looping and continue with checking threshold values one less or one more than the current threshold value.

Search for Optimal Threshold Values

Step 2

After finding a legitimate solution, adjacent threshold values in the range of current threshold ± Δ to determine a legitimate solution with a higher reward rate.

When no adjacent threshold returns a higher reward rate, the CAC algorithm returns the optimal value.

Numeric Data and Analysis

A Simulated Wireless Cellular Network with a Wrap-Around Structure.

Each cell having 6 adjacent cells The system with 1024 mobile user

roaming there cells Poisson distribution to model call

arrivals Exponential distribution to model

the duration of a call Call inter-arrival and call departure

rate to each mobile user by using uniform distribution

Performance Analysis

Parameters used in Simulation Study

Reward Rate versus Number of Mobile Units.

Partitioning CAC performs the worst in terms of reward rate among the four algorithms evaluated. up to 563 mobile users.

Threshold-based and spillover CAC algorithms generates legitimate solutions up to 768

Elastic threshold-based CAC is up to 819

QoS of admission Control Alogrithms.

Partitioning CAC rejects about 1% of low priority calls when the number of mobile user is 358, 0.7%, 2%, 3%, and 4.5% of class 1 handoff calls, class 1 new calls, class 2 handoff calls, and class 2 new calls, respectively.

Threshold-based and spillover CAC algorithms have similar performance characteristics.

Elastic threshold-based CAC is able to limit the acceptance rate of low-priority new calls to 92% by properly adjusting the low and high threshold values of low-priority new calls.

Conclusion

Elastic threshold-based CAC algorithm, even in heavy load condition, is still capability of satisfying QoS requirement.

Capable of leveraging the low threshold to regulate traffic and the

high threshold to reject traffic generated by service calls.

New threshold-based CAC algorithm has a goal to maximize the system reward rate with QoS guarantees.

Thank You !

Questions?