Optimizing Single-Cell Voice over WiFi Communications Through Cross-Layering Analysis

1. Performance Optimization of Single-Cell Voice over WiFi Communications Using Quantitative Cross-Layering Analysis Fabrizio Granelli (UniTN) Dzmitry Kliazovich (UniTN) Jie Hui (Intel Corp.) Michael Devetsikiotis (NCSU) June 19th, 2007

2. Motivation • Layering • Enable fast development of interoperable systems, but… • … limited the performance of the overall architecture, due to the lack of coordination among protocols • Cross-Layering • A novel design principle, whose idea is to allow coordination, interaction and joint design of protocols crossing different layers • Seems appropriate for specific scenarios, such as wireless, where independent layer design may be sub-optimal • No formal (quantitative) characterization of the cross-layer interaction among different levels of the protocol stack is available yet

3. Objectives and Contribution • Objectives • to identify and formalize the interactions crossing the layers of the standardized protocol stack; • to systematically study cross-layer effects in terms of quantitative models; • to support the design of cross-layering techniques for optimizing network performance; • to define design principles of Call Admission Control (CAC) strategies • Contribution • a general quantitative approach • methodological contribution: adopt “metamodeling” • a case study of VoWiFi cell: VoIP capacity and operator revenues optimization

4. Protocol Design Issues • Layering (ISO/OSI) • It is possible to model the ISO/OSI layer N entity as an object characterized • parameters of the object, pN • measurements that it can perform, mN • Cross-Layering • Weak Cross-Layering • interaction among layers of the protocol stack • includes “non-adjacent” interactions • Strong Cross-Layering • allows joint design of the algorithms within any entity at any level of the protocol stack • individual features related to the different layers can be lost due to the cross-layering optimization

5. Quantifying Cross-Layering • Quantifying the effect of potential cross-layer interactions is very important • to systematically relate such interactions to system outcomes • to quantify the decision to take such interactions into account • We propose to quantify cross layer interactions by defining factors (parameters) and effects (measurements) across layers • in a way that is common in system science and operations research

6. Quantifying Cross-Layering(cont.) • A system is characterized by • “factors” (controllable parameters) • “effects” (performance metrics) • The sensitivity of the system response and interactions can be captured using partial derivatives:

7. Quantifying Cross-Layering(cont.) • Using such tools, it is possible to optimize the performance ei with respect to a subset of pTOT under general constraints • by using steepest ascent, stochastic approximation, ridge analysis, stationary points, etc. • or to make local steps or decisions at a given operating point • in the context of game-theoretic or other economic-driven adjustments • or one may wish to dynamically control the response fk over time (optimal control)

8. Quantifying Cross-Layering(cont.) • The quantitative degree of cross-layer interaction and sensitivity will also guide one to a decision of whether to actually take a specific interaction into account or not • cross layer designs have implicit disadvantages in terms of cost and complexity • Some researchers have underlined that cross-layer design should be considered under a cautionary perspective [*] • a concept that our proposed framework integrates and rationalizes. [*] V. Kawada, and P.R. Kumar, “A Cautionary Perspective on Cross-Layer Design,” IEEE Wireless Communications, Vol. 12, No. 1, pp. 3-11, Feb. 2005.

9. Economic Considerations • Utility • “raw” performance metrics ei will typically be further incorporated into utility functions U(e) • express better how valuable the performance metric is to the system owner or user • examples include functions of the system throughput, overall delay or jitter, and system capacity • the utility function can have several forms and shapes • Prices • controllable parameters (factors or resources) will also likely to have actual (literal) or virtual prices, say $a per unit of design parameter X and $b per unit of Y

10. System Design Issues • System Design & Optimization • analytical, numerical or simulation-based methods could be used to achieve the design goals, either up front (i.e., parameter optimization), or on-line (i.e., optimal control) • More in detail, by employing the proposed framework, it is possible to select: • the optimal operating point of the system (direct consequence of the optimization process); • the proper cross-layer interactions to enable (based on sensitivity of the system); • the proper signaling architecture to employ (allowing to identify the set of parameters and measurements to use).

11. Case Study: VoWiFi Capacity • Network Model • Problem Statement: maximum # of VoIP calls, supported in an infrastructure Wi-Fi, with satisfactory QoS performance • Network Model: Infrastructure, N stations APP: G711 VoIP 64 Kb/s RTP/UDP/IP: header MAC: DCF with no RTS/CTS PHY: 802.11b, 11Mbps • Cross-Layer interactions: Between PHY, MAC, and APP

12. Case Study: VoWiFi Capacity • Inputs • X=[ DataRate ErrorRate NumofRetr VoicePktIntvl ] • Outputs • maximum # of VoIP calls supported by WLAN cell Y = N* with satisfactory quality • Constrains • Objective: acceptable voice quality (MOS = 3) • End-to-end delay measured between unpacketized voice: < 100 ms • Voice frame error rate: < 5% • Design Parameters

13. 40 Analysis 35 30 * 25 Metamodel VoIP Capacity N 20 15 10 Simulation 5 0 0 2 4 6 100 90 80 8 70 60 DataTxRate (Mbps) 50 10 40 30 20 12 VoicePktIntvl (ms) 10 Case Study: VoWiFi Capacity(cont.) • Choose and Fit the Metamodel • Second order polynomial RSM with interactions (R2=0.81) • Evaluate the Metamodel: comparison • Analysis > Metamodel > Simulation

14. Case Study: VoWiFi Capacity(cont.) • Metamodel properties • Maximum of N*(D, I, R, PER) corresponds to • 20 VoIP calls for D=11 Mb/s, I=70 ms, R=5, PER=10-9 For low rates (1 or 2 Mb/s) further retransmissions start to degrade system performance Model is not sensitive to low PERs Violates E2E delay threshold of 100 ms

15. Case Study: VoWiFi Capacity(cont.) • Cross-Layer Sensitivity and Performance Optimization • System is sensitive • Voice packet interval (I) and Packet Error Rate (PER) • System is less sensitive • Data rate (D) and Number of MAC layer retransmissions (R)

16. Case Study: VoWiFi Capacity(cont.) • Service Provider Perspective • Utility function: Pcall - Price charged for a single call Ppower - Marginal cost of a unit of transmitted power Dwasted - Bandwidth wasted for retransmission in packets/second Pcall / Ppower - chosen to be equal to 100 which corresponds to a policy to charge $1 per VoIP call while the price paid for power resouce is just ¢1 • Maximum revenues: • $18.89 with D=11 Mb/s, I=70 ms, R=5, PER=10-9 Operator revenues on per-call basis Resources required by retransmissions Resources required to maintain a certain data and error rates

17. Case Study: VoWiFi Capacity(cont.) • Mobile Terminal Perspective • Objective: long battery life while providing acceptable call performance • Main parameters • transmission data rate D • maximum number of retransmissions R • Utility function: where and relative weight against costs

18. Case Study: VoWiFi Capacity(cont.) • Design Principles • limitation on the number of active nodes, and thus a proper Call Admission Control (CAC), is required • overall system performance depend on many parameters which can be recognized and quantified at different layers • This motivates an introduction of CAC schemes which exploit metamodel information to provide proper cross-layer parameter setting for run-time system optimization

19. Conclusions • A general formal framework for • analyzing and quantifying cross-layer interactions and • supporting the design of cross-layering techniques to optimize network performance, including cost-benefit considerations • A case study from IEEE 802.11 VoIP is analyzed • From VoIP capacity, network operator and mobile terminal perspectives • Ongoing work • to test the proposed framework in more complex scenario • to provide guidelines in definition of high-performance cross-layering solutions • to design metamodel-based Call Admission Control (CAC) approaches

20. Thank you!