Transcript: New from BookNet Canada for 2024: BNC CataList - Tech Forum 2024
OSGi Community Event 2010 - Predictability vs Dynamism - Managing dynamic real-time applications
1. João Américo and Walter Rudametkin
Bull S.A.S./LIG Grenoble
Predictability vs. Dynamism:
managing dynamic real-time
applications
2. Outline
• Context
• State-of-the-art
• Problem Identification
• Suggested Approach
• Limitations
• Conclusions and perspectives
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 2/18
3. About
• Walter RUDAMETKIN
– PhD student at Université de Grenoble
• João AMÉRICO
– PhD student at Université de Grenoble
– BSc at UFRGS (2010), MSc at Université Joseph
Fourier (2010), Engineer Degree at ENSIMAG
(2009)
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 3/18
5. State-of-the-art
RTSJ: Real-time Specification for Java
• Issues: garbage collection, dynamic class
loading, thread scheduling, etc.
Dynamic Evolution/Adaptation
• Architecture modification at runtime
Real-time dynamic adaptive software
• Based on QoS objects (QoSkets), modes
(SOFA-HI/Blue-ArX), and real-time
adaptations for CCM (CIAO, Cardamom).
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 5/18
6. State-of-the-art
Real-time OSGi
• Works focused mainly on isolation issues:
ARFLEX Project, [Richardson, 2009],
AONIX’s Real-time OSGi model
• Industry initiatives: Oracle/BEA’s
WebLogic Real-time, Integration between
Perc and mBS
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 6/18
7. Problem Identification
• OSGi platform is inappropriate for
real-time applications
– Memory issues
– Scheduling issues
– Isolation issues
– Runtime software evolution
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 7/18
8. Simple Use Case
Update/Reconfiguration
Security
Camera
TFrame = 4 ms
Security
Camera
TFrame = 5ms
Motion Detection
System
Real-time
∑TFrame ≤ 10ms
Security
Camera
TFrame = 3ms
Security
Camera
TFrame = 6ms
Display
Application
Non real-time
Notation
Required Service
Provided Service
getFrame()
getFrame()
getFrame()
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 8/18
9. Suggested Approach
• Distinction between critical and non-critical
code
– Architecture freezing policy
– Dynamic Real-time SLA
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 9/18
10. Architecture Freezing
• Application = set of states
– Each state corresponds to an architecture
(service bindings)
State S2 State S3
Add
Remove
State S1
Update Update Update
Add
Remove
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 10/18
11. Architecture Freezing
• Real-time processing states
– Architecture modifications forbidden
State S2 State S3
Add
Remove
Add
Remove
State S1
Update Update Update
State RTS1 State RTS2 State RTS3
Enter RT state
Leave RT state
Enter RT state
Leave RT state
Enter RT state
Leave RT state
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 11/18
12. Service Level Agreement
Service Registry
Contract
Monitor
SLA
Needs
!
Notation
Required Service
Provided Service
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 12/18
13. Real-Time Dynamic SLA
• Extension to the D-SLA model [Touseau, 2010]
– Task type
– Period
– Worst case execution time (WCET)
– Resource Utilization
– Priority
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 13/18
14. Implementation
• iPOJO component model extension
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 14/18
15. Validation
Architectures frozen during
real-time processing states
SLM not implemented
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 15/18
16. Limitations
• One real-time application at a time
• Unknown update times
• Component characterization
– Resource utilization measures
• Overhead
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 16/18
17. Results
• Architectural Freezing solves:
– Dynamic update
– Service interruptions
•but not disappearance of physical devices
• Dynamic RT-SLA solves:
– Service admission
•based on resource consumption,
deadlines, …
• Both require modifying apps (explicit
notifications)
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 17/18
18. THANK YOU FOR YOUR
ATTENTION!
Contact: {Joao.Americo, Walter.Rudametkin}@imag.fr
September 2010 AMÉRICO, RUDAMETKIN – Predictability vs. Dynamism: Managing dynamic real-time applications 18/18