A System for Simulation, Emulation, and Deployment of Heterogeneous Sensor Networks

1
A System for Simulation, Emulation,and
Deployment ofHeterogeneous Sensor Networks

• Presented at SenSys 2004
• Baltimore, MD
• Lewis Girod, Thanos Stathopoulos, Nithya
  Ramanthan
• Jeremy Elson, Deborah Estrin,
• Eric Osterweil and Tom Schoellhammer
• CENS Systems Lab

2
Importance of Heterogeneous Systems

• Many ENS systems combine the strengths of mote
  and microserver platforms
• Motes perform simple sensing tasks with long
  battery life
• Microservers expend more energy to host
  sophisticated sensors and algorithms

Robotic nodes host cameras, tend clusters of
static motes
Clusters of dynamically taskable static motes
report data to nearest sink
Static motes report data back to fixed gateway
3
Heterogeneous Distributed Systems

• Vision complex application integrated from small
  and large nodes
• Integrates two already complex distributed
  systems in a non-trivial way
• Each system depends on input from the other
• To Build and Debug such applications, we must
• Integrate the system software running on the
  motes and microservers
• Drive simulations of microserver system with
  realistic input from mote net
• How can this be done?

Motes augment a microserver-based triggered
imaging app
4
Contributions of this Work
Emulation Array

• EmTOS
• Simple yet powerful mechanism for building and
  evaluating heterogeneous systems
• Run TinyOS code on EmStar microservers
• EmStar systems can participate directly in TinyOS
  network
• Enables range of simulation, emulation, and
  hybrid modes
• Uniform set of tools ? no cost to transitions
  among modes
• Experiments
• Characterize EmTOS accuracy
• Case Study
• Development of ESS using EmTOS

Mote in JR
EmView
5
Motes and Microservers
Mote (e.g. Mica2dot) Applications use TinyOS
component framework, are written in NesC. Native
TinyOS simulation via TOSSIM.
Microserver (e.g. Stargate) Applications run
above Linux, within the EmStar component
framework.Native EmStar simulation via EmSim.
So, how can we

• Integrate mote and microserver nets w/o
  reimplementing protocols?
• Simulate a complete system, made up of motes and
  microservers?

6
Solution EmTOS Wrapper Library
EmTOS Wrapper Library
Enables NesC applications to provide new EmStar
services
tos/leds
tos/eeprom
tos/tasks
User defined
User defined
EmStatusServer
EmPacketServer
TOS status
Unmodified NesC Application
Wraps an unmodified NesC app into an EmStar
module
SenseToRFM
TimerC
AM
Implements TinyOS API and low-level components
LEDs
EEPROM
UART
ADC
RadioCRCPacket
ClockC
Underlying EmStar Services
By connecting to existing EmStar services
link/mote0
motenic
This fairly small investment of effort -- the
wrapper library and some TinyOS components that
make up a new TinyOS platform -- simultaneously
addresses both goals of heterogeneous simulation
and integration!
Transceiver (Mote)
7
Enables Mote-Microserver Integration
Example ESS
ESS network

• NesC-based Multihop tree routing protocol
• Runs natively on motes
• Runs in EmTOS wrapper on microserver
• Wrapped version exposes EmStar devices
• Used by microserver sink implementation
• Increased visibility on microserver

ESS
EssDse
Multihop
link/mote0
motenic
Dse
TimerC
AM
Transceiver (Mica2)
RadioCRCPacket
ClockC
ADC
Mote RF Channel
8
Enables Heterogeneous Simulation

• EmTOS makes TinyOS look like an EmStar module
• Thus EmSim can simulate them like any other
  EmStar node!
• EmSim runs N EmStar nodes in parallel on a
  centralized server
• Simulated microservers run natively (runs real
  code)
• Simulated motes run TinyOS/NesC app in EmTOS
  wrapper

Example ESS
9
Requirements for Development Tools

• Deployment Target
• Large Scale
• Works in Reality
• Limited Visibility
• No Out Of Band debug channel
• Key Simulation Requirements
• Real Code for rapid iteration
• Common world model for all nodes
• Visualization, logging, analysis tools
• Work across many modes simulation, emulation,
  deployment

10
EmSim Modes
(Runs at scale of emulation array)
Pure Sim
Deployed
11
Pure Simulation Mode
Simulation Server

• Pure Simulation Mode
• All Application code runs on server
• Maximum Visibility
• Full logs and status
• Minimum Reality
• Modeled mote RF channel / MAC
• Emulated motes timing
• Initial debug and regression testing

Modeled Mote RF Channel

Motes
Microservers
12
Emulation Array
Simulation Server
Serial MUX
HostMote Serial Protocol

• Emulation Array
• Used for Emulation/Real/Hybrid Mode
• Real motes installed in environment
• Serial multiplexer connects server to real motes,
  replaces channel model
• Limits scale improves reality

Motes
Microservers
13
Emulation Mode
Simulation Server
Serial MUX
HostMote Serial Protocol

• Emulation Mode
• All Application code runs on server
• Same visibility as simulation
• Real mote RF channel
• Motes run Transceiver packet proxy
• Packet timing

Motes
Microservers
14
Real Mode
Simulation Server
Serial MUX
HostMote Serial Protocol

• Real Mode
• Microserver code runs on server
• Mote code pushed to physical motes
• Full visibility for microservers
• Motes report only partial status info
• Stub applications gateway to tools

Dbg Stub
Dbg Stub
Dbg Stub

Motes
Microservers
15
Hybrid Mode
Simulation Server
Serial MUX
HostMote Serial Protocol

• Hybrid Mode
• Mixture of real and emulated motes
• Range of possibilities

Dbg Stub
Dbg Stub

Motes
Microservers
16
EmTOS Emulation Accuracy
VS

• EmTOS enables integration and simulation
• But it does not present the same timing as a real
  mote.
• How can we characterize the differences and
  assess their importance?

• Performed two experiments
• Compared timer jitter in EmTOS vs. Mica2
• Compared packet latency in Emulation Mode vs.
  Mica2

17
Mica2 Timer Jitter is Application Dependent
2 Timers compute Task
1 Timer
3 Timers
10µs
1µs
10ms

• Compared requested time against actual time of
  event
• Real motes exhibit widely varying,
  application-dependent jitter
• Depends on number of tasks scheduled before the
  Timer task
• EmTOS exhibits consistent 10ms jitter resulting
  from jiffy timer (1ms jitter for Linux 2.6
  Kernels)

18
Emulated Packet Timing is Delayed
Distribution of RTT
60ms
60ms
180ms
120ms
EmTOS
Mica2

• Native motes access channel directly while EmTOS
  must traverse a serial line to access the RF
  channel via Transceiver
• Results in similar timing distribution, with 60ms
  excess RTT

19
Do the differences matter?

• EmTOS cant do precise timing and hardware
  interrupts
• e.g. EmTOS can emulate many network layer
  protocols
• EmTOS cant emulate the TinyOS MAC layer
• Validation is easy using EmTOS and an Emulation
  Array
• Compare runs in Emulation Mode and Real Mode
• Emulation Mode timing, interrupts by EmTOS
• Real Mode timing, interrupts by Mica2s in
  emulation array
• Validation catches design flaws masked by
  emulation mode
• e.g. Early version of ESS
• Interest messages were flooded too quickly
• Emulation delay improved performance
• But still caught pre-deployment!

20
Case Study ESS Development

• Development process
• Use of simulation, emulation, and real modes
• Real mode used to validate vs. Emulation mode
• Analysis and performance measurement
• Real mode packet traces with precise timing
• Offline analysis of collisions and retx
• In deployment
• EmTOS integrates Multihop tree-building and
  transport protocol with microserver software
• EmStar provides visibility and robustness for
  microserver code
• Live topology data returned via transport layer
  piped to visualization tools

Pure Sim
Deployed
21
Current and Future Work

• Current
• CENS Contaminant Transport Project
• Based on ESS codebase, but closes loop locally to
  control robot
• TinyDB with heterogeneous routing layer (Intel
  Research)
• Uses EmTOS to run TinyDB code on Stargate, and
  attract compute-intensive query processing to
  more powerful nodes
• Future
• Triggered Imaging and Acoustic Processing
• Inputs from mote systems trigger microserver
  applications
• Mote Herding Further work on mote-microserver
  integration
• Stargate-side middleware for applications across
  mote clusters
• e.g. function handoff, resource discovery,
  reliable data transport

22
Conclusion

• EmTOS is a simple, powerful technique for
  simulating, emulating, and deploying
  heterogeneous sensor systems
• EmTOS facilitates
• Integration of motes and microservers by enabling
  code to be shared across platforms
• Simulation of heterogeneous systems composed of
  motes and microservers within a single world
  model
• While EmTOS is not an accurate emulation of a
  Mica2
• Its close enough for development and integration
  of many application- and network-level components

23
Thank You

• More information
• The EmStar website, for downloads, documentation,
  mailing lists http//cvs.cens.ucla.edu/emstar
• A System for Simulation, Emulation, and
  Deployment of Heterogeneous Sensor Networks, in
  SenSys 2004 http//lecs.cs.ucla.edu/girod/papers
  /emtos-sensys04.pdf
• EmStar a Software Environment for Developing
  and Deploying Wireless Sensor Networks, in
  USENIX 2004 http//lecs.cs.ucla.edu/girod/papers
  /emstar-usenix04.pdf