Network traffic engineering

Transcription

1 University of Roma Sapienza DIET Network traffic engineering Lecturer: Andrea Baiocchi DIET - University of Roma Sapienza andrea.baiocchi@uniroma1.it URL: Lecture 1 Introduction to network traffic engineering a.y. 2014/2015 Traffic engineering: what is it?!2 Traffic engineering deals with the application of probability and optimization theories to analysis and design of service systems. Analysis Calculate/estimate performance Design Evaluation of which resources to use and how much of them. Service systems Input-output function with given quality specs Examples: protocol, router, server, network

2 Network Traffic Engineering!3 Applied to networking From isolated network elements to protocols, to networks Service system abstraction Allows generale results, applicable to different domains e.g., transportation networks, logistics, power distribution networks, supply chains and inventory management, workflow systems Methodologies are applicable to different problems e.g., Markov chains and processes, renewal processes, queueing theory, convex optimization, graph theory NTE target!4 The aim of NTE is to build mathematical models, given a service system specification: Structure, policies, intrinsic service capacity Statistical description of demand service request arrivals amount of required service The model gives mathematical rules to tie together: DEMAND (USERS) SYSTEM CAPACITY (RESOURCES) PERFORMANCE METRICS (QUALITY)

3 Example!5 A base station of a cellular network, covering a single cell Demand assignment resource management, with equal rate communication channels assigned for a whole session Arrivals? Service? Appropriate model? Performance metrics? Application objective!6 The final objective is to design service system that meet given service grade constraints at a minimum cost (minimization of required resources) for the given/ expected demand. To that purpose, we need: Methods to quantify and forecast demand (service characterization, measurements); Method to evaluate the system capacity (e.g., models of system, lab measurements, filed trials, simulations); Methods to evaluate the service grade offered by a given configuration of the system with given demand Methods for recovery actions, to restore the capabilities of the systems after congestion and or mulfunctioning.

4 Example (follows)!7 Minimize number of channels for given set-up failure probability Hand-offs: how can we deal with them? Demand quantification: mobility pattern, service preferences, session duration More complex if more flexibility allowed: different rates, priorities Evaluation of service grade: Erlang model Recovery actions: some channels are turned-off; the base station fails (e.g., power failure) Erlang loss model usage!8

5 The art of modeling!9 From system description to a mathematical model Key: simplifying assumptions and hypotheses Requires: In-depth technical knowledge of system working and context; Mathematical skills to adopt the most useful models. A system model can be analyzed by Theoretical means (e.g., queueing theory) Computer aided simulations Lab/field experiments Example (follows)!10 Aim: optimize use of spectrum Full knowledge of radio resource management MAC protocol or multiple access scheme Types of service, requirements. Radio channel modeling (attenutation, shadowing, fading) Physical layer modeling (e.g., Shannon limit capacity) Full knowledge of user behavior patterns Mobility Traffic demand Desired QoE/QoS, willingness to pay

6 NTE approaches pros and cons!11 APPROACH Theoretical model grossly simplified gain insight Simulation model Lab/field trial laborious, costly very expensive, inflexible, costly many details accounted for the ultimate word Modeling approach!12 Observation Model Analysis Results Check

7 NTE tasks!13 NTE tools!14

8 Example!15 Let us now work out an example: delay equalization of a packet stream Applications: streaming audio/video Technical approach: Introduce additional, variable delay at receiving end Issues: Late packets Receiver buffer sizing