Desiderata

1
Desiderata

• (Slides to accompany lecture
• these are not understandable on their own)

2
Operating System

• An Operating System is like a government
• protection from within (police)
• virtual memory, file system
• Protection from without (army)
• ssh, passwords, ..
• A Pervasive Computing OS is like the European
  Union
• like a government but respect autonomy of members

3
Adaptive, Distributed Strategy

• No solution presented today
• No problem presented today
• Some examples for parallel/dist that I worked on
  in past I like
• Hope that the lesson will translate to other
  challenges in pervasive comp.

4
Local Decisions, Global Opts

• Should you rent or buy skis?
• 50 to rent
• 300 to buy
• At first rent and eventually buy
• but when?

5
Amortized Cost Analysis

• Without any knowledge of future
• rent 6 times, then buy
• never pay more than 600
• twice the optimal
• Not such a big win when done once
• but if follow this for everything then is a big
  savings
• Similar analysis when know something about the
  future (prob)

6
Load Balancing

• Assumptions
• all tasks can be executed anywhere
• tasks are compute intensive
• multiple time quantums
• there are P processors, all connected
• Load Balance solution
• Centralized periodically tell centralized server
  and get/give tasks
• Distributed periodically balance with another
  processor

7
Adaptive Balancing

• Hard to choose the right time period
• too frequent, wasted overhead
• too rare, too many idle processors
• If processor has N tasks, then
• load balance every 1/N context switches
• choose proc at random and balance load
• always send tasks at end of queue
• implm each proc has a load balance task
• just gets execute like any other task

8
Critical Section

• Classic Algorithm

a true turn a while ( turn a b )yield
critical section a false
b true turn b while ( turn b a )
yield critical section b false
9
Snapshot Rollback
10
Checkpoint
11
Adaptive snapshot/rollback
12
Conclusion

• there are deep research issues