Elastic Thresholdbased Admission Control for Qo S Satisfaction

Elastic Threshold-based Admission Control for Qo. S Satisfaction in Wireless Networks with Reward Optimization for Multiple Priority Classes April 6, 2010 M. Conlan A. Moini

Content Background Key Qo. S metrics for wireless cellular networks Call Admission Control (CAC) algorithms Elastic threshold-based CAC algorithm System model Performance model Analysis results Comparison with other CAC algorithms Conclusions References

Background � Mobile wireless networks must increasingly carry multiple classes of services with distinct Quality of Service (Qo. S) requirements • real-time multimedia services Standard voice calls Streaming video/audio • non-real-time services SMS text messages Picture mail Email � Network providers need a method for optimizing the cumulative value of services they provided � This presentation focuses on a threshold-based CAC algorithm which determines optimal threshold levels maximizing system “reward” while satisfying Qo. S constraints for multiple priority service classes

Key Qo. S Performance Metrics Cellular Wireless Network � Blocking probability of new calls � Dropping probability of handoff calls • a handoff occurs when a mobile user with an ongoing connection leaves current cell and enters another cell • an ongoing connection may be dropped during a handoff due to unavailability of wireless channels (insufficient bandwidth in new cell) Qo. S • Constraints observed blocking probability should be less than the blocking probability threshold service i Bi for ≤ Bteach Bi ≤ Bti class i h h n n You can reduce handoff-call-drop probability by rejecting new connection requests, thus increasing in new-call blocking probability.

Call Admission Control (CAC) � Mechanism to regulate traffic volume in (wireless or wired) networks • intended to ensure, or maintain, a certain level of quality of service • work by regulating total utilized bandwidth, total number of calls, packets or data bits passing a specific point per unit time • Extensively studied for single-class network traffic, such as voice (realtime) � Threshold-based • when a defined limit is reached or exceeded, new calls may be prohibited from entering the network until at least one current call terminates or prevent new calls from entering the network only if the resources of a particular type would be overburdened • example: keep the dropping probability of handoff calls and/or the blocking probability of new calls lower than pre-specified thresholds � Partition-based algorithms • partition system resources and allocate distinct partitioned resources to serve distinct service classes � Priority-based • regulation of calls according to priority descriptors � Graceful • degradation service quality of individual calls can deteriorate to a certain extent before new calls are denied entry

Call Admission Control (CAC) Algorithms � Threshold-based Algorithm • Ogbonmwan, Li and Kazakos (2005) • 3 threshold levels for a system with two service classes • used to reserve channels for voice handoff calls, new voice calls, and data handoff calls • threshold values are periodically reevaluated based on workload conditions � Distributed CAC algorithm • Haung and Ho (2002) • partitions channel resources in each cell into three partitions: �real-time calls partition �non-real-time calls partition, and �shared partition used by both classes calls to share • applies a threshold value to new calls to satisfy more stringent Qo. S requirements for handoff calls • uses an iterative algorithm to estimate call arrival rates to each cell in the heterogeneous networks

CAC Algorithms (cont. ) � Bandwidth Reservation and Reconfiguration • Ye, Hou and Papavassilliou (2002) • mechanism to facilitate handoff processes for multiple services Common Characteristics of CAC Algorithms Call admission decisions based on meeting or not exceeding a certain threshold levels • Example: keep dropping probability of handoff calls and/or blocking probability of new calls lower than pre-specified thresholds Handle Qo. S requirements without considering “value” issues associated with service classes, i. e. , what value priority service classes will bring to a system

Threshold-based CAC Algorithm Chen and Chen (2006) Assigns distinct, discrete thresholds to each service type Shares all available channels among all service classes to achieve higher utilization Leverages thresholds to limit traffic from low-priority calls, hence reserving more bandwidth for highpriority calls Limitations: • suffers from use of discrete thresholds which cuts traffic from service classes abruptly and reject any further traffic • How to select “appropriate” threshold level

Elastic Threshold-based CAC Algorithm for Multiple Service Classes with Priorities � Extends earlier work by Chen et al. (2006) � Utilizes two thresholds for each service class i: � low threshold � high threshold � Rejects a fraction of class i new service calls when low threshold is reached � Rejects all class i new service calls once high threshold is reached

CLAIM: Elastic Threshold-based CAC Algorithm produces optimal results! By allowing multiple service call types to share all channels and by limiting call arrivals of low-priority service classes, elastic threshold-based CAC algorithm produces optimal results: � maximizes systems reward while meeting Qo. S requirements “reward” refers to any kind of “value” brought to the system due to services example: “revenue” � generates higher rewards compared to existing CACs

Network Reward Function (assuming 2 -priority service classes*) reward earned from servicing class i handoff calls per unit time reward earned from servicing class i new calls per unit time *: extensible to multiple service classes without loss of generality

Service Qo. S Requirements (assuming 2 -priority service classes) Qo. S constraints are expressed in terms of blocking probability thresholds: Observed handoff dropping probability and new call blocking probability of class i generated by a CAC algorithm must not exceed the corresponding threshold probabilities. Blocking probability threshold for handoff calls Blocking probability threshold for new calls

System Model (Wireless Cellular Network) � Each cell has C channels where C can vary depending on the available bandwidth in that cell � When a call of service class i enters a handoff area from a neighboring cell, a handoff call request is generated � Threshold is reached if accepting an incoming call will cause the number of channels used to exceed the threshold value. Each service call has its specific Qo. S requirement • dictated by its service type attribute (e. g. , real-time, non real-time) • requires certain number of bandwidth channels • imposes system-wide Qo. S requirements

Elastic Threshold-based CAC Algorithm for Multiple Service Classes with Priorities new call class i low threshold System high threshold handoff call class i low threshold high threshold rejects a fraction of class i new calls when is reached and rejects all class i new calls when is reached starts blocking a fraction of class i handoff calls when is reached and blocks all class i handoff calls when is passed.

Elastic Threshold-based CAC Algorithm for 2 -priority Service Classes* *: extensible to multiple service classes without loss of generality

Elastic Threshold-based CAC Algorithm for 2 -priority Service Classes Low threshold is triggered if a new low-priority class 2 call arrives when the number of channels used by the system is greater than by. CAC then starts rejecting a fraction of (class 2) call arrivals until a class 2 a new call arrival causes the number of channels being used exceed the high threshold Once the high threshold of new calls is reached, the system rejects all class 2 new calls. Similar behavior for class 2 handoff calls

Elastic Threshold-based CAC Call Admission Probabilities n : total number of channels allocated in the system Prob. of accepting a new call of service class i Prob. of accepting a hand-off call of service class i ki : number of channels required by a service call

SPN Model for Elastic Threshold-based CAC

SPN Model for Elastic Threshold-based CAC Places Transitions * :

SPN Model for Elastic Threshold CAC Transitions Enabling Predicates : :

SPN Model for Elastic Threshold CAC Arrival Rates if = if 0 if is disabled.

SPN Model Parameters Blocking/dropping probabilities as a function of arrival rate: Reward earned per unit time, per cell reward earned from servicing class i handoff calls per unit time reward earned from servicing class i new calls per unit time i V : assigned reward per call for service class i (no distinction between new and handoff calls)

Finding Optimal Threshold Combination Challenge: find a set of threshold levels that provide “legitimate” solution Two-step process • Step I : finding a “legitimate” solution • Step II: determining a locally optimal solution by applying a greedy search starting from the legitimate solution found in Step I Finding a “legitimate” solution • Method I : set all thresholds at max capacity (C) and incrementally reduce low threshold, in reverse priority order, until “legitimate” solution is found • Method II: start with all thresholds set to minimum channel size required to support the Qo. S constraints and incrementally increase until “legitimate” solution is found (invoked only if 1 st method fails). Next perturb threshold levels using a greedy search algorithm to optimize reward while satisfying Qo. S requirements adjacent threshold levels (current threshold ) for values with higher reward, if any. Check legitimate solution : maximizes reward per unit time while satisfying Qo. S constraints

Comparison of Elastic Threshold-based with other CAC Algorithms Model and analyze wireless cellular network performance using simulation Apply competing CAC algorithms to measure system Qo. S and reward rate performance • threshold-based • partition • spillover • elastic threshold-based Consider two distinct priority service classes • real-time (e. g. video) and non real-time (e. g. voice) • each service type requires a number of bandwidth channels to satisfy its bandwidth Qo. S requirement • handoff calls have a higher priority than new calls since disconnection of an ongoing call is considered very undesirable

Simulated Wireless Cellular Network with Wrap-around Structure • • • 6 adjacent cells 1024 users Random destination Random speed Random pause time

Simulation Parameters new call blocking probability handoff call blocking probability for each service class i (i =1, 2)

Reward Rate vs. Number of Mobile Units Elastic threshold-based CAC algorithm produced highest reward.

Qo. S of Call Admission Algorithms Elastic threshold-based CAC algorithm ensures Qo. S for more users.

Conclusions � Elastic threshold-based CAC algorithm is superior • satisfies Qo. S requirements even in heavy load conditions • generates high rewards despite increased traffic generated by high population • leverages low threshold to regulate traffic (rejecting just a fraction of traffic) and the high threshold to reject traffic generated by service calls • outperforms existing CAC algorithms for Qo. S satisfaction and reward optimization • is extensible to multiple priority service classes � Threshold-based and spillover CAC algorithms perform reasonably well under moderate load � Partitioning CAC algorithms perform poorly among all

References 1. S. E. Ogbonmwan, W. Li, D. Kazakos, Multi-threshold bandwidth reservation scheme of an integrated voice/data wireless network, in: 2005 International Conference on Wireless Networks, Communications and Mobile Computing, Maui, Hawaii, June 2005, pp. 226– 231. 2. Y. -R. Haung, J. -M. Ho, Distributed call admission control for a heterogeneous PCS network, IEEE Transactions on Computers 51 (2002), 1400– 1409. 3. J. Ye, J. Hou, S. Papavassilliou, A comprehensive resource management for next generation wireless networks, IEEE Transactions on Mobile Computing 1 (4) (2002) 249– 263. 4. I. R. Chen, C. M. Chen, Threshold-based admission control policies for multimedia servers, The Computer Journal 39 (9) (1996) 757– 766. 5. O. Yilmaz and I. R. Chen, "Elastic threshold-based admission control for Qo. S satisfaction with reward optimization for servicing multiple priority classes in wireless networks, “ Information Processing Letters, Vol. 109, No. 15, July 2009, pp. 868 -875.