Analysis of Two-Layer Performance Models by Using Generalized Approaches from Teletraffic Theory L. N. Popova Institute for Mobile Communications, University of Erlangen-Nürnberg, Germany V. B. Iversen COM - DTU, Technical University of Denmark, Lyngby, Denmark

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 2 Motivation  Simultaneous support for a wide range of services with different characteristics on a common carrier  Universal frequency reuse High spectrum efficiency  Flexible service bit rate  Statistical Multiplexing among services Higher resource utilization Problem:  Diversity of service requirements difficult to ensure sufficient QoS for all services Multi-Service UMTS Network:

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 3 Capacity of UMTS  UMTC cell capacity is power based capacity  Each new user increases the total power level in the frequency channel  Capacity is limited by the amount of interference in the air interface  Packet user doesn’t occupy a channel continuously (on-off source ) produce interference only when it is actually transmitting no fixed value of the maximal cell capacity Inherent but uncontrolled Statistical Multiplexing

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 4 Conventional Traffic Models  Classical traffic models from fixed networks only deal with the system behavior at connection level (call arrival process)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 5 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 6 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 7 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 8 Aggregated Traffic Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 9 Overbooking Adjusted according to the measured quality Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 10 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 11 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 12 Conventional Traffic Models

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 13 Connection Level Call Admission Control Connection Level Call Admission Control II f not enough capacity block call completely EE rlang BCC model HH ard blocking only Blocked-Call-Cleared: too pessimistic performance results

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 14  Classical traffic models don’t take into account diversity of service requirements on packet-oriented wireless networks  Just a rough approximation for network performance  Don’ t ensure sufficient QoS for all services  Important features of WCDMA radio interface are neglected: –Variable on –off user transmission rate –Packet buffering/delay –Dynamic cell capacity (wireless interference)  Handover strategies are not included Problem Definition Problem Definition

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 15 Contribution  Proposal of a new unified analytical traffic model for a system with mixed services: –Analysis of interaction between the two performance levels: Connection level call admission procedure Packet level call handling process –Individual performance assessment for each service class  Proposal of an extended call handling policy buffer scheme  Analysis of impact of buffer on: –The average system performance –Individual users’ throughput

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 16 Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 17 Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 18 Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 19 Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 20 Aggregated Traffic Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 21 Overbooking Adjusted according to the measured quality Multi-Layer Traffic Model (recap)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 22  In case of overbooking, there is loss for the period of congestion  Fry-Molina BCH model  Loss sharing between simultaneously transmitted blocks of distinct users  Soft Capacity Multi-Layer Traffic Model (recap) Blocked-Call-Held: The same TTI during decoding: all blocks are affected

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 23 Modified Multi-Layer Traffic Model Blocked-Call-Buffered

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 24 Modified Multi-Layer Traffic Model Blocked-Call-Buffered

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 25 Modified Multi-Layer Traffic Model Congested Traffic Blocked-Call-Buffered

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 26 Modified Multi-Layer Traffic Model Carried Delayed Traffic Buffered Traffic

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 27 Advantages of Suggested Approach (1) Main difference between two models is the system throughput utilization Blocked-Call-Held:Blocked-Call-Buffered:

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 28  Fast and stable algorithm  Supports the coupling between connection and packet- level QoS characteristics.  Individual performance measure for each traffic stream  More realistic model for analysing behaviour of multiple traffic flows  Supports all classical loss/delay models.  Applicable to performance prediction and the optimum design of virtually arbitrary networks Key aspects of the algorithm: Advantages of Suggested Approach (2)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 29 Traffic Model and Assumptions  Radio interface of W-CDMA, Perfect Power control, Uplink  Multiple service classes with different QoS-parameters: Binomial – Poisson – Pascal (BPP) multi-rate traffic  Connection - Level: Blocked-Call-Cleared (Erlang) model: –Each traffic stream is characterized by: Mean offered traffic Peakedness Data rate required by stream  Packet – Level: Blocked–Call–Buffered model –On-off traffic with activity factor –State-dependent loss probability Neighbor cell interference (log normal)

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 30 Numerical results (1)  Blocked-Call-Held model Packet Handling Policy: Setup:  mixed services: Radio stream (Poisson); = 4(8)128 (Erl.) Web browsing (Engset); = 40 (Erl.) (Pascal); = 37(Erl.)  Activity factor: 0.5; 0.3; 0.7  FIFO-principle  No trunk reservation  Total cell capacity N=128 channels

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 31 Numerical results (1)  Blocked-Call-Held model  Blocked-Call-Buffered model Packet Handling Policy: Setup:  mixed services: Radio stream (Poisson); = 4(8)128 (Erl.) Web browsing (Engset); = 40 (Erl.) (Pascal); = 37(Erl.)  Activity factor: 0.5; 0.3; 0.7  FIFO-principle  No trunk reservation  Total cell capacity N=128 channels

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 32 Modified Call Handling Policy Packet Handling Policy:  B Blocked-Call-Held model  B Blocked-Call-Buffered model  B Blocked-Call-Buffered model with wireless interference Setup:  mixed services: Radio stream (Poisson); = 4(8)128 (Erl.) Web browsing (Engset); = 40 (Erl.) (Pascal); = 37(Erl.)  Activity factor: 0.5; 0.3; 0.7  Other cell-interference factor=0.55  Total cell capacity N=128 channels

Larissa Popova and Villy Iversen: Analysis of Two-Level Performance Models 33 Conclusions  Goal: Analysis of resource allocation scheme for large-scale queuing networks with WCDMA radio interface  Proposal of an extended call handling policy (introducing of buffer scheme)  How: by using a novel generalized algorithm from extended teletraffic theory: –Combines properties of both loss and queuing systems – Addresses connection and packet-level QoS metrics simultaneously. –Allows explicit theoretical analysis of complex multimedia traffic behavior –Generalizes scheduling policy and service priority discipline for performance evaluation of traffic streams with different QoS provisioning problems.  Findings: Comparison of Blocked-Call-Held scheme with Blocked-Call- Buffered traffic model: –Blocked-Call-Buffered traffic model is a more realistic approach for modelling system processes on the complex network packet-level –Applicable to performance prediction and the optimum design of virtually arbitrary networks