BenchMANET: MANET Reference Configurations

	Thomas Kunz
	Carleton University
	(joint work with Mohamed Abou El Saoud and Samy Mahmoud)

Motivation

	Service Discovery in MANET
		Allows automatic discovery and location of services
		Essential for pervasive computing scenario
		MANETs levy additional unique challenges:
			Network Topology constantly changing
			Multi-hop
			Decentralized
			Limited battery and processing power
	Need Comprehensive and Realistic Performance Evaluation Framework to evaluate and compare SDP/MANET protocols.

BENCHManet Evolution – Tests

	Classified MANET applications, and derived 10 benchmark tests. Each test represents an application class:
		Business & Commercial
			Low Mobility - Collaborative Conference (conf)
			High Mobility - Event coverage (event)
		Crisis Management
			Almost-Static Target - Rescue Operation (rescue)
			Moving Target - Police Pursuits
		Personal Area Networks (pan)
		Residential Mesh (mesh)
		Vehicle Applications
			Vehicle-Roadside Networking (vr)
			Vehicle-Passenger Networking (vp)
		Military Applications
			Organized motion - Soldiers Marching (march)
			Unexpected continuous motion - Combat (combat)

BENCHManet Evolution – Mobility

	Assigned suitable mobility model and parameters to each of the 10 benchmark tests.
	For instance, in a military march:
		Movement parallel to travel
			Column Mobility Model
		Human Speed
			1m/s ± 1m/s
		No Pauses
			0s ± 0s

BENCHManet Evolution –
Other Configurations

	Each of the 10 tests possess other scenario-specific features. The following parameters were chosen for each test:
		Spatial Area
		Network Size
		Server Agent Ratio
		User Agent Ratio
		Services Per Node
		Service URL Advertisement lifetime
		Number of Simultaneous Requests
		Server Duplication Ratio

Overview of SLP

	UAs: mediate for users/applications
		Request Service information (active)
		Listen to service broadcasts from DAs (passive)
	SAs: represent services
		Reply to SrvRqsts (active)
		Broadcast service information to DAs (passive)
	DAs: intermediate centralized service brokers
		Cache service information from SAs
		Satisfy UA requests accordingly

SLPManet – Adapted SLP

	SLPManet implements all required features in SLP specification (RFC2608).
	Optional features not suitable for MANET:
		DAs: can not have pre/continual-existing nodes.
		Authentication Blocks: digital signatures for security, not goal of research.
		Other optional messages: add complexity and consume scarce resources.
	Designed for small multi-hop MANETs under cooperative administrative control.
	Provided additional caching of service requests/reply

Evaluation – Experimental Setup

	SAs and UAs picked randomly from [0,NS)
	Simulation Time = 2000s
	Service Requests
		Total per scenario = 1000
		Inter-arrival = exponential [1000s, 2000s), mean=1360s
		Each UA requests ~1000/UAs services
		Each request is for a random service type
	BENCHManet test suit run 10 times with different movement files.

Evaluation – SLPManet (1)

	Overall Discovery Success
		9/10: > 95%
		combat: 50% due to troublesome mobility and topology
	Service Lookup Latency
		Peak latency high for scenarios where nodes are further distant apart, resulting in many retransmissions before a SrvRply is successfully received

Evaluation – SLPManet (2)

	Bandwidth Consumption
		mesh: large number of duplicated services and large number of nodes causing longer delays and more retransmissions of SrvRplys

Evaluation – Improvement (1)

	Overall Discovery Success
		9/10:   0% – 1%
		combat:   35%
	Service Lookup Latency
		Average:   24% – 98%
		Peak:   0% – 82%

Evaluation – Improvement (2)

	Bandwidth Consumption
		Average:   15% – 97%
		Peak:   -8% – 58%
		vr: deterioration not statistically significant
	Aggregate Bandwidth
		16% – 96%

Summary of Contributions

	Provided Taxonomy of MANET applications.
	Provided BENCHManet, a practical comprehensive performance evaluation framework for MANET protocols.
	Evaluated performance of SLPManet using BENCHManet.
	Proposed and implemented a simple extended caching modification to SLPManet.
	Evaluated performance gain of improvement.

Future Research

	Compare current active-discovery SLPManet with passive-discovery versions.
	Compare SLPManet against other SDPs.
	Comparison between application-level service discovery and cross-layer service discovery proposals.
	Enhance BENCHManet by using more sophisticated mobility models.