Patterns for Resource Reallocation in OODRE Systems

1
Patterns forResource Reallocationin OODRE
Systems

• Joe Crossjoseph.k.cross_at_lmco.com

Patrick Lardieripatrick.j.lardieri_at_lmco.com
This work was sponsored by DARPA under contract
number F33615-01-C-1847
2
Problem and Context

• These patterns apply to systems
• on which the service requirements1 change, or
• in which the set of available resources2 changes
• Hence resources must be allocated to different
  purposes at different times
• How and when are such resource reallocations
• planned?
• triggered?
• implemented?

1. E.g., cycles per second for processing, bandwidth
   or reliability for inter-process communication
2. E.g., processors, switches, radios

3
Two Solution Patterns
ProactiveResource Reallocation
ReactiveResource Reallocation
4
Known Uses
Reactive Resource Reallocation
A web page server that reacts to changesin
hits-per-minute
Proactive Resource Reallocation
An air traffic control system that has a normal
mode and a degraded mode
5
How to Choose

• Reactive resource reallocation is good where

• QoS requirements
• loads
• component availability

all change slowly relative to the time neededto
plan and implement a reallocation.

• Proactive resource reallocation is good when you
  have hard time bounds on reallocations.

6
Common Features
Both proactive and reactive resource reallocation

• Require reflective capabilities for use
  inlong-lived systems
• Require, in general, reflective capabilities that
  can forecast performance (e.g., under ltthesegt
  loads, ltthisgt performance would be provided)
• Raise concerns of convergence and stability
• Present challenges when Powerpointing a skeptical
  customer (cf. the Tacoma Narrows Bridge)

7
Interaction in ProactiveResource Reallocation