Sample document annotated with topics:

Realistic Large Scale Online Network Simulation Xin Liu and Andrew Chien, In Proceedings of the ACM Conference on High Performance Computing and Networking, SC2004, Pittsburgh , Pennsylvania, November 2004.

Words in this document have been colored by the following topics:

[ communication network routing ] network host routing packet router link traffic path internet nodes attack address
[ networked computer systems ] server system client resource request service resources network access services
[ parallel processing ] processor parallel data performance communication memory computation thread program
[ graph partitioning algorithms ] partitioning partition cut local cost bisection search gain algorithm heuristic

Note: Words not assigned to the top six topics are in black, and stopwords are in gray. Words have been tokenized (punctuation removed and changed to lowercase).

Realistic Large-Scale Online Network Simulation

Xin Liu and Andrew A. Chien
Department of Computer Science and Engineering
and Center for Networked Systems
University of California, San Diego
{xinliu, achien}@cs.ucsd.edu

Abstract

Large-scale network simulation important technique studying dynamic behavior network network protocol emerging classes distributed application grid peer_peer etc large_scale realism critical requirement network simulation grid application studies work here extend previous effort three key way first study network larger previous studies router second scale study realistic network structures bgp4 ospf routing versus flat ospf routing finally describe evaluate new profile based load_balancing approach called hierarchical profile based load balance extensive large scaleexperiment withprofile based load balance prof flat routed ospf network show prof outperform several techniques based topology static application information result those multi network motivate invention new hierarchical technique hprof cluster network nodes achieve desired minimum link latency mll key determinant simulation parallelism applies graph partitioner hprof explicitly control tradeoff simulation efficiency available parallelism producing robust superior performance large_scale network including both singleas multi network hprof improve load imbalance reduce simulation time about router simulation executed node cluster parallel efficiency achieved simulation over providing substantial capabilities simulating large network summary advances demonstrate realistic large_scale network simulation network router comparable large tier isp network like accomplished system

1. Introduction

Historically network simulation emulation extensively explore behavior network protocol because difficulty modeling application behavior detail most simulation simple application model exercise protocol network advent large application tightly couple compute storage network techniques study resources together emerging particular large_scale network simulation important technique studying dynamic behavior network network protocol emerging classes distributed application including peer_peer grid application network important contributor application performance application generate large amount network traffic overall application performance critical wide variety simulation system built model network behavior based discrete event simulation

massf network simulation tool key component microgrid system built group ucsd study dynamic behavior grid application microgrid enables execution complete grid distributed application key requirement network simulator targeted large_scale study application resource infrastructures

first requirement scale internet scale network many network simulation project massf utilizes cluster system achieve scalable performance harnessing scalable compute resources massf system user application together themselves interesting distributed application load balance network simulation itself key problem scalability previous work formulated load balance problem graph partitioning problem applied classical graph partition algorithm solve three approaches exploiting topology topology application placement profile based presented evaluated moderate sized network result showed exploiting static topology application placement information improves load balance profile based approach further improves load balance achieved paper improve new hierarchical approach evaluate them much larger network

second requirement large_scale network simulation simulate detail structure realistic network previous published work massf addresses simulation accuracy validation paper address issue realistic network topology routing selection while much research explores realistic internet like topology generator background traffic few effort explore realistic network routing most large_scale simulation pursuing shortest_path routing ospf well known large multi network routing amongst different domain controlled bgp policy routing therefore connectivity does equal reachability realistic internet network simulation support bgp routing among autonomous system reasonable bgp routing policy configuration massf support detailed bgp4 routing protocol here address remaining problem how generate reasonable bgp routing policy large network

paper demonstrate techniques enable realistic large_scale online network simulation techniques together make realistic large_scale simulation study network coupled application performance possible specific contribution paper include modeling network resources information services enable user grid researcher grid operator study arbitrary collection resources network addition microgrid virtualizes transparently allowing direct study complex application middleware whose internal dynamic difficult model accurately real application software middleware unchanged executed arbitrary virtual grid structures short microgrid provides virtual grid infrastructure enables scientific systematic experimentation dynamic resource management techniques adaptive application supporting controllable repeatable observable experiment because rate execution component system application network etc controlled wide range relative performance system combination modeled microgrid virtual grid agentnodes input queue evaluating previous load_balancing techniques top prof online simulation network times larger router study realistic network routing structures bgp4 ospf routing versus flat ospf routing describing evaluating new load_balancing approach called hierarchical profile based load balance hprof developing set heuristic automatic realistic bgp routing configuration improvement internet like topology generation evaluating range load_balancing techniques top htop prof hprof simulation both single multi network demonstrate hprof improve load imbalance reduce simulation time router simulation executed node cluster demonstrate realistic large_scale network simulation network router comparable large tier isp network like accomplished massf system output queue network modeling ssf api live traffic thread pool simulation engine virtual real mapping server wrap socket grid application

remainder paper organized follow section provides background microgrid massf internet hierarchy section describes load balance approaches scalability challenges summarizing existing partition algorithm presenting hierarchical partition approach larger scale network experiment section demonstrate scalability partition approaches single network section first introduce set heuristic rules automatic bgp routing configuration provide evaluation result load balance approaches against internet like multi network realistic bgp routing result discussed along related work section finally section summarizes work point future direction research