Hybrid Simulation Testbed

1
Hybrid Simulation Testbed

• Rajive Bagrodia
• Junlan Zhou, Zhengrong Ji, Mineo Takai
• Parallel Computing Lab, UCLA

2
Hybrid Simulation Testbed

• Objective
• To fulfill quest for network modeling tools that
  can
• Seamlessly Interconnect with physical networks
• Interact with real applications/operating systems
  at
• Application / Transport / Network / MAC layer
• Run as fast as / faster than real time
• Mimic large-scale next generation wireless
  networks

3
Our Approach
Application
User space
Transport
Kernel space
Network
Network
Media access
Media access
Device driver
Device driver
Software
Network device
Network device
Hardware
Media
Media
Physical Compoents
Simulated Compoents
4
Past Studies

• Hybrid simulation
• Integrating packet capturing libraries into
  existing simulation tools
• Modeling cross-interaction between operational
  application and underlying network protocols
• Scalability
• Network emulation
• Imitate hosts in wireless networks using regular
  PCs and Ethernet devices
• Intercept packets from network layer to device
  driver to emulate dynamic network conditions
• Limited scalability

5
Our Framework

• System overview

• Emulated Nodes
• Each mapped to a physical host
• Running operational application
• Simulated Nodes
• Multiple of them are mapped to a physical host
  and simulated by a Qualnet instance
• Generating background traffic

Emulated node
Emulated node
Emulated node
Simulated subnet
Emulated node
6
Our Framework

• Implementations (1)

Application
Transport
Qualnet
IP
Application
Transport
Virtual device driver
Network
Mobility trace
MAC model
MAC
PHY
Channel model
Radio device model
Device driver
Device driver
Network device
Network device
LAN
7
Our Framework

• Implementations (2)

• Emulated Nodes
• MAC model
• Simulate behaviors of MAC protocol
• Channel model
• Mobility trace
• Generate node movement
• Radio device model
• Simulate device operation (carrier sensing,
  backoff, signal reception)

IP
Virtual device driver
Data
MAC model
Mobility trace
Node movement
Incoming/Outgoing Pkt, channel sensing status
Pathloss, Fading,
Radio device model
Channel model
RTS, CTS, Data, or ACK
Device driver
8
Our Framework

• Implementations (3)

• Simulated Nodes
• incoming traffic is intercepted at device driver
  and injected into PHY layer of Qualnet.
• traffic generated by the nodes simulated locally
  is broadcasted to nodes emulated or simulated
  remotely.

Qualnet
Application
Transport
Network
MAC
PHY
RTS, CTS, Data, or ACK
Device driver
9
Our Progress

• Completion of hybrid testbed design
• Incorporation of high resolution timer in Linux
  kernel
• Improving scalability of wireless network
  simulation for hybrid testbed
• Development of first hybrid testbed in progress

10
Improving Scalability of Wireless Network
Simulation

• Objectives
• Preserve accuracy
• Improve network simulation efficiency
• As a foundation for integration of detailed radio
  and channel models
• hybrid simulation testbed

11
What is the overhead in wireless network
simulation?

• Propagation model
• Signal has long distance of reachability
• Multiple interferences
• accumulation of weak signals
• Physical device model (802.11)
• CSMA/CA
• BO timer
• SINR
• Common approach is to drop signals weaker than
  carrier sensing threshold (CST), i.e. to limit
  signals reachability

12
Misleading results of commonapproach
Experiment setup 100 nodes 2000x2000m2 AODV 30
random CBR sessions 512-byte pkts 210
pkts/sec Same traffic load for all sessions
13
Our approach

• Applying better distance limit

D?2500m with these parameters
Table. Common experiment parameters
14
Validation of D

• Experiment setup
• 400 nodes uniformly distributed
• 4000x4000m2 terrain
• AODV
• 120 CBR sessions between random pair of nodes and
  8 pkts/sec each
• 512 bytes / packet
• Portion () of sessions are one hop traffic

15
Validation of D (cont)
16
Performance Evaluation of D

• Experiment setup
• Nodes are uniformly distributed
• 30 of nodes have a CBR session to another node
  within two hop distance
• 10 pkts/sec 512 bytes/pkt
• Varying network size
• Varying node density
• Varying traffic load

17
Varying network size
18
Varying other parameters
Varying node density
Varying Traffic Load
19
Reduce PHY layer events

• In common network simulators
• A packet transmission by PHY will generate a
  signal arrival signal end event
• Needed to update SINR CCA
• Event scheduling overhead is enormous
• Solution
• Lazy Event Scheduling
• Corrective Retrospection
• (LSCR)

20
Lazy Event Scheduling

• No events scheduled for non-receivable signals

21
Corrective Retrospection

• Delayed packet error evaluation

22
Corrective Retrospection (cont)

• Timer corrections
• Keep adjusting timers remaining time until
  actual remaining time and projected remaining
  time converges.

23
Events reduction by LSCR (per sec in network
simulation)
24
Speedup of LSCR
Additional speedup of LSCR 2.5?4.2 as
network size grows