Transversal Issues in Real-Time Sense-and-Respond Systems

1
Transversal Issues in Real-Time Sense-and-Respond
Systems
Ahmad T. Al-Hammouri Vincenzo Liberatore Huthaifa
A. Al-Omari Case Western Reserve
University Stephen M. Phillips Arizona State
University Support by NASA NAG3-2578,
NAG3-2799, NNC05CB20C, NNC04AA12A
NSF CCR-0329910 Department of Commerce TOP
39-60-04003.
2
Real-Time SR

• Objective
• Tele-epistemological vision
• Networked SR extend human reach beyond spatial
  barriers
• Remote control of physical environment by sense
  and respond
• Potential Applications
• Industrial automation
• Automatic asset mgmt. (RFID)
• Disaster Recovery
• UAVs LNK03
• Home robotics

3
Our contributions

• Formulate Real-Time SR for Networks research
• Middleware demonstration
• Playback buffers
• Congestion control
• Work in-progress
• Multihoming, implementation, etc.

4
(No Transcript)
5
Real-Time SR and Networks

• Real-Time SR involve the physical environments
• Real-Time objectives and constraints
• Real-time objectives
• Correctness, reliability, and safety
• Flexibility, scalability, and efficiency
• Communication networks
• No timeliness or QoS guarantees BW limitation,
  losses, delays, and jitter
• Focus has been on
• Control theory, real-time systems, middleware,
  and data bases

6
Main thesis

• Networks research is critical for Real-Time SR
• Playback buffers
• Congestion control
• Quality-of-Service
• Major technology gap
• Multimedia streaming, anyone?

7
Transversal issues

• SR control environments
• Radically differ in complexity and in
  applications
• Include hierarchical levels of abstraction
• Transversal issues
• Address the necessity of providing QoS for SR
• These are
• Adaptability and tolerance
• Congestion control
• Security, fault tolerance, etc.

8
Transversal issues

• We are currently involved in _at_ Case
• Adaptability and tolerance
• Congestion control

9
Adaptability and tolerance

• SR systems must be adaptable to
• Delays
• Jitter

10
Adaptability and tolerance (cont.)

• .di s vary randomly
• Decompose di as di t ?i
• .t the nominal round-trip delay (predictable
  across i)
• .?i the variation (jitter) in every round-trip
  time delay
• Two subproblems
• Predictable delays
• Jitter

11
Adaptability and tolerance (cont.)

• Predictable delays (?i 0, di t)
• Methods to mitigate predictable delays
• Sacrificing performance to preserve stability,
  e.g., robots teleoperation
• Use of system models predict a future state
• Jitter
• Jitter causes inaccurate predictions in
  delay-compensation techniques
• Playback buffers can eliminate jitter

12
Playback buffers
Packet
Sequence number
generation
Adapted from Peterson Davie
T
ime
13
Playback buffers multimedia vs. SR

• Different performance metrics
• Round-trip delay jitter
• Playback delays are determined by controller
• Controller is away from the host that applies the
  signal
• Controller needs RTT to detect a delay spike

VoIP stream
14
Our contribution

• Combined strategies of playback, control laws,
  and adaptive sampling
• Controller determines playback delay based on RTT
  samples
• Adaptive playback delay
• Based on the playback delay, Controller
  determines
• control signal, and
• sampling rate

15
Transversal issues

• We are currently involved in _at_ Case
• Adaptability and tolerance
• Congestion control

16
Congestion control

• Match sources transmission rates to network
  capacity
• Achieve fair bandwidth allocation between
  different flows
• Example
• S1 r1, r2, r3 0.5, 1.5, 0.5
• S2 r1, r2, r3 0.25, 1.75, 0.75
• S3 r1, r2, r3 0.0, 2.0, 1.0

17
Congestion control (cont.)

• Define a Utility function, Ui(ri)
• Performance of system i transmitting at rate ri
• Mathematical formulation
• max S U(ri)
• s.t. S ri Cl , l 1,, L
• and ri rmin,i

18
Congestion control (cont.)
queuen
Sink
Source
19
Our contribution

• Choice of a utility function, Ui(Ai, ri)
• Function of speed of physical dynamics, Ai
• Monotonically increasing and strictly concave
  with ri
• Tailoring the general optimization
  framework for SR

20
Additional directions

• Implement these methods
• Middleware vs. lower-level layers
• Exploit transport layer protocols
• SCTP Multihoming and reliability to SR
• Power management control

21
Thank you

• Comments
• Questions