Exploration of Real time Embedded system Architectures

1
Exploration of Real time Embedded system
Architectures

• Architectural solutions for enhancing the
  real-time behavior of distributed
• embedded systems
• A Beowulf-class architecture proposal for
  real-time embedded vision

Presented by Srikanth
2
The common thread

• COTS
• Distributed/Parallel

The differences

• OS overhead
• Explored Architectures
• Proposed Architecture

• Classic ES issues perf, pow, vol, prog
• Novel Parallel Arch.
• Exploiting diff levels of parallelism

3
Architectural solutions for enhancing the
real-time behavior of distributedembedded systems

• Addresses a common problem in conventional
  architectures undesirable timing variations in
  applications temporal behavior
• Cause overload caused by operating system
  activities when dealing with concurrent and
  time-triggered processes.

4
Problem background

• RTOS
• a key component of real-time distributed embedded
  systems
• eases the task of programming concurrent
  applications
• However
• RTOS usually compete for the same CPUs with other
  application tasks
• can (and usually do) influence the overall
  systems performance, leading to non
  deterministic temporal behavior

5
and approach

• a hardware and software architecture
• dedicated hardwares using COTS
• uClinux and some real-time extensions
• Scheduling partitioned, allocated to a special
  processor
• minimize RTOS administrative overhead
• achieve a deterministic temporal behavior

6
Multi-processor Architecture for Embedded
Real-time Systems
7
Spring Architecture for Real-Time Systems

• distinct processors for the application and for
  the rest of the system
• optic communication channels
• a VLSI device to perform the scheduling
  activities
• Most of its tools and APIs were proprietary what
  negatively impacts portability and programming
  aspects

8
Architectural Support for Predictability in Hard
Real Time Systems

• a multi-processor architecture consisting of one
  or more general-purpose microprocessors and one
  coprocessor for running operating system tasks
• to enhance systems predictability
• improve systems performance by minimizing the
  number of interrupts to operating systems task
  execution
• A programming language was specifically developed
  for the target hardware

9
Proposed Architecture
10
A Beowulf-class architecture proposal for
real-time embedded vision

• a new type of parallel computer architecture
  dedicated to real-time vision
• architecture is based upon the Beowulf concept
• But, considers the constraints of embedded vision
• on the fly processing of video streams,
• small volume,
• and low power consumption

11
What is Beowulf

• a multi-computer,
• Scalable architecture consisting of MCOTS
• Runs freely available OS and software
• inter-connected via a
• high-bandwidth network like Fast Ethernet,
  Giganet, SCI,Myrinet or ATM

12
Architectures for real-time embedded vision

• High performance
• Scalability
• Resistance to obsolescence
• Volume and power consumption
• Easy to use and efficient mechanism for video
  data acquisition
• Cost
• Programmability

13
Proposed Architecture
14
Computing nodes

• Power PC G4
• Altivec vector processing unit
• Useful for vision applications
• Very good Gflops/Watt ratio
• very small motherboard (16x19 cm)

15
three-layered software architecture

• The operating system is Linux
• The low-level parallel programming model - MPI
• portable communication library
• The high-level parallel programming model - based
  upon algorithmic skeletons

16
(No Transcript)
17
Time for results
18
(No Transcript)
19
(No Transcript)
20
(No Transcript)