1 / 42

Performance Optimization Problems at Various Layers of Telecommunication Networks

Performance Optimization Problems at Various Layers of Telecommunication Networks. Overview of Research G. R. Dattatreya datta@utdallas.edu (Papers at www.utdallas.edu/~datta/papers). Top level list. Physical layer Noise and data estimation in WIreless LANs MAC layer

ryder
Télécharger la présentation

Performance Optimization Problems at Various Layers of Telecommunication Networks

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Performance Optimization Problems at Various Layers of Telecommunication Networks Overview of Research G. R. Dattatreyadatta@utdallas.edu (Papers at www.utdallas.edu/~datta/papers)

  2. Top level list • Physical layer • Noise and data estimation in WIreless LANs • MAC layer • Modeling for better channel utilization • Network layer • Bursty traffic through static and Ad Hoc networks • TCP/IP traffic • Application layer project – Alcatel VoIP system • Industry level projects – QuEST Forum and SBC

  3. Salient features of research approach • Each problem is motivated by and formulated for real world current technology applications • Probability Theory and Stochastic Models for problem formulation and analysis • Analytical solutions, with justified sequence of approximations, if necessary • Demonstration through implementation of solutions, simulation, and analysis of simulation results

  4. Physical layer – Wireless LANs(Larry Singh) • Noise affects BER • So does data statistics • Both are also time varying • Figures for noise and data parameters needed for minimum BER receiver design • They should be estimated on-line – as the receiver operates • Elegant, easy to implement solution • Theretical analysis of performance of solution

  5. Wireless LANs (contd.) • Systems: PAM with AGC and Coherent Detection • Open problem: Estimation of AGC parameter • After problem formulation, no need for EE background • Slides from WCNC presentation follow

  6. MAC layer problems • Overall motivation: • Optimal channel utilization through distributed actions by individual transmitters • Approach: • Identify control parameters for each transmitter • Measure observable channel characteristics • Tune the control parameters • Think globally; act locally

  7. MAC problem: Some details • Start with a transmitter protocol • Identify control parameters: • parameters of random wait, p-sensing, etc. • Approximate random variables to lead to Markov chain: • Even a contant 52 micro second is turned into an Exp. rv. With a mean of 52 micro second !! • Try to develop the composite Markov chain for multiple transmitters.

  8. MAC problem (contd.) • Appears to be a very large Markov chain !! (K. Srinivasan) • Try to simplify • even with crude approximations • Try to express overall performance as a function of transmitter control variables: • Assume same values for a control variable of all transmitters

  9. MAC problem (contd.) • Formidable problem • Some groundwork is complete • Other approaches: • Start with modest systems and simpler protocols (only one control variable) • Try simulating with one transmitter searchnig for optimal solution while others use constant control parameters • Extend to distributed searching by all

  10. Network layer research • Mobile ad hoc network performance (S. Kulkarni) • Bursty traffic engineering (S. Kulkarni, M. Qiu) • modeling, generation, analysis, queue and network performance • Cross layer issues: • Using TCP model in Network traffic model • Combining network traffic control with Datalink control -- Joint with Dr. Saquib, EE

  11. Routing in ad hoc networks (S. Kulkarni) Need an algorithm that • Has low overheads • Lowers end-to-end packet delay • “Adapts” to changes in • traffic pattern • topology • Uses multiple routes, if available = load balancing

  12. Routing Strategy • Distributed approach • No global knowledge of topology or traffic • Each node “learns” independently • Local routing decisions

  13. Routing Strategy (contd.) • Probabilistic routing • Nondeterministic path from source (Si) to destination (Sj) • Load balancing using multiple routes • Avoid congestive hot spots • Packets not forwarded to already visited neighbors • Unacknowledged packet delivery

  14. The Algorithm: Statistically Multiplexed Adaptive Routing Technique (SMART)

  15. SMART: Packet Loss Comparison(Static Topology) Poisson Traffic Multiplexed Pareto Traffic

  16. Experiments With Mobility(Dynamic Topology, Self-Similar Traffic) • Random waypoint mobility model • 30 nodes in 1.5km x 2.5km field • Transmitter range 600m • Random destination, nodes move in straight line • Velocity uniformly distributed between 1m/s and some maximum • Nodes do not pause during simulation • Traffic: Self-similar (100-source muxed Pareto) as before

  17. SMART: Packet Delivery Ratio(Dynamic Topology, Multiplexed Pareto Traffic)

  18. Some indicative figures Queue saturation • Poisson: 98-99% load • Bursty traffic: 30-80% load Network saturation • Poisson: load >70-80% • Multiplexed-Pareto: >40-50% Adaptive algorithm reduces losses in static network by • 25% for Poisson traffic • 50% for multiplexed-Pareto traffic Mobility: Delivery figures (datagram) at moderate loads • 80% packet delivery at low speeds • >60% packet delivery at high speeds

  19. Bursty traffic engineering • Origin of burstiness: • Heavy tailed distributions for various traffic aspects • Simplest models: M/G/1 and G/M/1 • M/G/1 leads to unbounded average queue length • Can G/M/1 model burstiness?

  20. G/M/1 model for bursty traffic queue (S. Kulkarni) • Inter-arrival times (IATs) with infinite variance • Very large IATS – is this a problem? • To make up for the modest average, • One large IAT also implies several very small IATs some times – so, bursty

  21. Effect of Traffic Burstiness Effective load Vs Long term load average, iid Pareto IATs

  22. Characterization of bursty traffic • Generation, measurement, estimation are difficult due to random variables with infinite variance • Unbounded variance also leads to “long range dependence (LRD)” • Alternative models that truncate this effect beyond significant levels are developed

  23. Alternative models of bursty traffic • High order autoregressive models • Useful in generation (S. Kulkarni) • Time series (M. Qiu) -- Traffic amounts over successive intervals -- Useful for analysis and parameter estimation • State transition diagrams (M. Qiu) -- Traffic amounts over successive intervals -- Useful for analysis and parameter estimation -- Analysis of TCP/IP traffic (M/G/infinity models) • MMPP (L. Singh) • An intermediate model between exponential and Pareto extremes • Also intermediate between iid and LRD • Parameter estimation in MMPP

  24. State transition diagrams • Poisson arrival of TCPs • Each TCP lasts for Pareto amount of time (due to unbounded variance of file sizes) • Each TCP pumps a Poisson sequence of packets for its duration • Packets from multiple TCPs are multiplexed into the router queue • The heavy tail effect vanishes !!

  25. State transition diagrams (contd.) • Results in a Markovian system • Two discrete variables: (N_tcp, N_pckt) • Unfortunately, non-product form !! • Two dimensional Z-transform • Partial differential equation • Work in progress (M. Qiu)

  26. MMPP model of bursty traffic • Lesson from state transition diagram: • Unbounded variance and Unbounded LRD sum may not be really required • Extended but limited dependability: • MMPP models: Arrival rate may fluctuate with rates controlled by a Markov chain

  27. Characterization of MMPP: Parameter estimation (Larry Singh) • Two stages: • Estimate the mixture of multiple exponential random variables • Formulated as an Optimization problem • Use earlier result to estimate parameters of controlling Markov chain • Sample results follow

  28. Application layer project • Performance of proxy server for VoIP • Alcatel Funded project • Modeling the functioning of the system (from performance point of view) • Analysis; identification of parameters that crucially affect performance • Researching to obtain figures for these parameters • Evaluation of performance figures

  29. Application layer project (contd.) • Important questions that needed answers • Can we shift the required facilities and processing burden from end terminals to a central server • How many end terminals can be supported per server, given a performance requirement • Average delay of call establishment • Chance of not getting service

  30. Recent industry level projects • QuEST Forum and SBC funded • Study of performance and quality of products and systems • Performance: • Figures on reliability, failure times, how well the failuers were dealth with, etc. • Hierarchical model of performance of large systems • Quality: • Hierarchical model from Customer Satisfaction Data

  31. Industry level projects (contd.) • Questions • How well are the performance and quality data correlated • Can we use their trends to identify problem areas for corrective actions

  32. Conclusion • Problems from current technologies • Mathematical modeling • Sequence of approximations • If necessary • as appropriate • Mathematical solutions • Implementation/Simulation • To demonstrate • Especially if approximations have been made • Assessment

More Related