Transparent Environment for Replicated Ravenscar Applications

1
Transparent Environment for Replicated Ravenscar
Applications

• Luís Miguel Pinho
• Francisco Vasques
• Ada-Europe 2002
• Vienna, Austria
• 18-20 June 2002

2
Summary

• Motivation
• System Model
• Replication Management Framework
• Repository of task interaction objects
• Example
• Conclusions

3
Motivation

• Computer Control Systems
• Are present in a wide range of application
  domains
• Are expected to perform correctly (value and
  time) even in the presence of faults
• Need to guarantee real-time and fault tolerance
  properties of applications
• Distributed systems

Replication as a means to achieve fault tolerance
4
Motivation

• New Challenges
• Commercial Off-The-Shelf Components (COTS)
• Minimise cost and time to market
• Lack of real-time and fault tolerance properties
• Requires software-based fault tolerance
  techniques
• Pre-emptive priority driven model
• Higher flexibility
• Non-deterministic
• Increasing complexity
• Manage real-time and fault tolerance requirements
  ...
• together with the controlled system
  requirements
• A transparent and generic solution is required

5
System Model

• Distributed fault-tolerant hard real-time
  applications
• Application environment
• Multitasking environment
• Guaranteed execution resources
• Replicated applications
• Tolerate COTS components faults
• Provides the sameenvironment in all nodes

6
System Model

• Defines a replication model
• Tasks are joined in components
• The component as the replication unit
• A component may be spread over several nodes, and
  several components can share a node
• De-coupling replication and distribution roles

7
Replication Management Framework

• Application development approach
• From the application programmer perspective,
  simple objects are available to share data and to
  release tasks
• Applications are developed without considering
  replication and distribution
• Application configuration is performed by object
  replacement
• Framework structure

8
Replication Management Framework

• Object Repository
• Mapping common task interaction in real-time
  systems
• Shared Data Objects for mutual exclusion
• Use of timed messages for replica determinism
• Release Event Objects for sporadic task release
• With/without data
• Asynchronous task communication
• Tasks can not block accessing remote data
• Remote objects are locally replicated
• All writes are atomically disseminated

9
Replication Management Framework

• Transparency
• Full transparency
• Inefficient
• Difficult to know off-line the characteristics of
  the application
• Solution
• The objects provide a transparent interface, by
  which application tasks are not aware of
  replication and distribution issues
• In a later configuration phase,
  distributed/replicated resources replace those
  simple resources
• Full characteristics of the application are known
  off-line

10
Object Repository

• Use of Generic Packages
• Reuse of implementation mechanisms
• Object parameterisation (configuration) at
  compile-time
• Same Interfaces (except for instantiation)
• Encapsulation
• Private implementation based in Protected Types
• Mutual exclusion
• Sporadic Tasks control

11
Object Repository

• Interfaces

12
Object Repository

• Interfaces

13
Object Repository

• Implementation

14
Application Example

• Simple Application

15
Application Example

• Application Code Objects

16
Application Example

• Application Code Tasks

17
Application Example

• Application Configuration

18
Application Example

• Application Configuration

19
Application Example

• Application Code Node 1

20
Application Example

• Object Instantiation before Configuration

21
Application Example

• Object Instantiation after Configuration

22
Conclusions

• A suitable framework for the development of
  fault-tolerant hard real-time applications
• Targeting
• Pre-emptive fixed priority applications
• COTS-based systems
• Transparency and genericity in application
  development
• Distribution and replication only considered in a
  later configuration phase
• Using the semi-transparent approach
  predictability is achieved

23
The End
Thank You