distributed monitoring of mesh networks
DESCRIPTION
Distributed Monitoring of Mesh Networks. Elizabeth Belding-Royer Mobility Management and Networking (MOMENT) Lab Dept. of Computer Science University of California, Santa Barbara Joint work with Krishna Ramachandran and Kevin Almeroth. Motivation: Monitoring. - PowerPoint PPT PresentationTRANSCRIPT
Distributed Monitoring of Mesh Networks
Elizabeth Belding-RoyerMobility Management and Networking (MOMENT) LabDept. of Computer ScienceUniversity of California, Santa Barbara
Joint work with Krishna Ramachandran and Kevin Almeroth
Motivation: Monitoring
crucial for robust network operation benefits to network operators, system designers,
researchers
essential for evolving network technologies critical last piece in the product conception-
design-development-improvement loop helps bridge the gap between the expected
(simulations) and the unexpected (real-world)
The Big Picture Deployment
UCSB 25 node mesh network (NSF WHYNET project)
Monitoring and Measurement (DAMON) UCSB mesh IETF meetings LocustWorld, IV deployments
11,000 AODV nodes in 50+ countries Simulation models
movement models traffic models AODV refinement
The Big Picture Deployment
UCSB 25 node mesh network (NSF WHYNET project)
Monitoring and Measurement (DAMON) UCSB mesh IETF meetings LocustWorld, IV deployments
11,000 AODV nodes in 50+ countries Simulation models
movement models traffic models AODV refinement
Outline
DAMON Design and Architecture DAMON Implementation DAMON@IETF Conclusions
Design Challenges
Device mobility Resource constraints Fluctuating link quality Short-lived network connections
Design Choices: Pervasiveness of Monitoring Solution
Strategy of using a centralized network element fails no hierarchical structure to
mobile networks mobility
Monitoring mobile networks requires pervasive solution nodes participate in
monitoring Amount of pervasiveness
complete coverage strategy limited coverage strategy
Pervasiveness
NetworkState
Pervasiveness tradeoffs
Design Choices: Pervasiveness of Monitoring Solution
Strategy of using a centralized network element fails no hierarchical structure to
mobile networks mobility
Monitoring mobile networks requires pervasive solution nodes participate in
monitoring Amount of pervasiveness
complete coverage strategy limited coverage strategy
Pervasiveness
AnalysisEffort
Pervasiveness tradeoffs
Additional Design Choices
Number of data sinks single sink? multiple sinks?
Temporal property of monitoring information determined by monitoring requirements classifications
time dependent information, e.g. topology information time independent information, e.g. packet logs
require differentiated handling of data
DAMON: Distributed Architecture for MONitoring mobile networks Overview
agents within network collect information information stored at sinkssink auto-discoveryresiliency to sink failures
Architecture
Agents within network send monitoring information to sinks
Sinks emanate periodic beacons facilitates auto-discovery and resiliency to
sink failures
Sink Auto-discovery
beacons contain agent instructions and hop count
agents use hop count to choose primary sink
Sink Auto-discovery Proximity-based
association (hop count)simple, low overheadbut, can lead to uneven
distribution of agents to sinks
Tradeoff between beaconing frequency and sink detection latency
Monitoring Information
Time dependent i.e., energy left on a device,
neighbors typically small in size packaged into time
dependent digests (TDDs) transmitted to sink
frequently unreliable transmission
Time independent i.e., packet logs, daily
traffic statistics typically large in size broken into small-sized
chunks called time independent digests (TIDs)
reliable transmission
Client Framework
Packet Classifier: categorizes packets based on types, dispatches to appropriate packet handler Beacon Listener: handles beacons TDD dispatcher: handles received TDDs Collectors: summarize routing table info or link quality estimates
in TDDs and TIDs
Packet Classifier
Collector1 Collectorn
Beacon Listener
TDD Dispatcher TID Dispatcher
File Server
Digest Classifier
…
Network
Client Framework
Digest Classifier: delivers digests created by Collectors to appropriate module TDD Dispatcher for immediate transmission to sink File Server for TIDs for later delivery to sink
TID Dispatcher: periodically retrieves digests for transmission to sink
Packet Classifier
Collector1 Collectorn
Beacon Listener
TDD Dispatcher TID Dispatcher
File Server
Digest Classifier
…
Network
DAMON Implementation Goals:
monitor ad hoc network behaviormonitor AODV performancemetrics of interest
throughput traffic distribution control packet overhead mobility patterns
Implementations for Linux and Microsoft Windows
DAMON Information Collection
AODV control packet summaries RREQ, RREP, RERR, Hello received packet counters UDP payload and timestamp
Topology data routing table deltas AODV-NEIGHBOR TDDs sent every minute
Data traffic statistics IP source and destination application protocol type packet size
DAMON@IETF 58th IETF Meeting in
Minneapolis, MN, November 9-14, 2003
Deployment goals: validate DAMON design track IETF topology evaluate AODV
performance observe traffic/mobility
patterns AODV Implementation
Linux, Windows (thanks Intel!)
130+ downloads 20+ simultaneous ad hoc
network members
Network configuration complete coverage
strategy one gateway provided
Internet connectivity to ad hoc network users
one sink deployed to collect information
ad hoc network co-located with 23 IETF APs
nodes used tool called PUDL to avoid unidirectional links
PUDL
Periodic Uni-Directional Link detector periodic unicast probes between each
neighbor pair sequence numbers used to measure
reliability under some threshold (40%), link filtered
from AODV
DAMON@IETF: Network Topology
Network Troubleshooting Connectivity problems with gateway reported
during 13:00-15:30 IETF session on November 11th
Node ID % Broadcast Hello
% Unicast Probes
1 91.8 74.1
2 76.26 12.69
3 92.06 36
4 74.73 42.18
5 69.23 54.1
6 95.42 11.4
7 97.85 6.66
Lessons from Connectivity Information
1. No correlation between reception of unicast and broadcast packets
2. Routing protocols should select routes based on how reliably a path delivers unicast packets
3. Relying on thresholds to avoid unidirectional links can eliminate links that are necessary for connectivity
Traffic Distribution
Per Protocol, With Link Filtering Per Protocol, Without Link Filtering
AODV Traffic Distribution
0
10
20
30
40
50
60
RREQ RREP RERR Hello
With LinkFiltering
Without LinkFiltering
Conclusions
Monitoring essential for robust network operation
DAMON overcomes challenges associated with mobile network monitoring
Future work: more DAMON deployments and analysis tools
http://moment.cs.ucsb.edu/DAMON
Funding provided by NSF and Intel Corporation