vignettes/overview/msd-paper-presentation.md

In vsimko/trafficmsd: Multiscale Decomposition of Network Traffic

Automated Extraction of Network Traffic Models Suitable for Performance Simulation

author: Viliam Simko date: 16.10.2015 autosize: true

Authors | Institutions --------------------|---------------------------------------- Viliam Simko | FZI Forschungszentrum Informatik Piotr Rygielski | Universität Würzburg Samuel Kounev | Universität Würzburg Klaus Schilling | Universität Würzburg Felix Sittner | ZfT Zentrum für Telematik Doris Aschenbrenner | ZfT Zentrum für Telematik

Context and Motivation

Nowadays data centers - Increasingly dynamic due to virtualization - High intra-data-center traffic (migration of VMs)

We focused on - Modeling performance of network in a data center - Network throughput - Assuming signals with some periodic component

Currently - Architecture derived automatically - Traffic models created manually

Requirements

• Derive automatically
• Decompose into generators / intervals
• Minimize the model size
• Preserve signal shape
• Tunable trade-off: precision / model size

High-level overview

TODO Image $\rightarrow$ Simulation

DML + DNI (Descartes Modeling Language)

High-level language → simulation models

Simulators: OMNeT++, 2 x QPN - Each simulator uses different abstractions - Each requires specific knowledge

Tradeoff: accuracy vs. simulation overhead

Related Work

Probabilistic models - TODO link - TODO link - TODO link

LIMBO - A load Intensity modeling platform

Typical traffic dumps

Typical traffic dumps - densities

Model of a simple traffic generator

$G = (scale, amplitude, begin, end)$

Parameter | Description --------------|------------------------------------------------- scale | Generator emits packets every $2^{scale}$ seconds. amplitude | Size of the packet to be emitted. begin | Beginning of the activity end | End of the activity

Decomposition into traffic generators

TODO Image

Example: 10 minutes of traffic - before optimization

Error in decompose(.) : time series has no or less than 2 periods

vsimko/trafficmsd documentation built on May 3, 2019, 7:08 p.m.