11
Distributed Process Manag ement 1 Learning Objectives • Distributed Scheduling Algorithms • Coordinator Elections • Orphan Processes

Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

  • View
    217

  • Download
    1

Embed Size (px)

Citation preview

Page 1: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 1

Learning Objectives

• Distributed Scheduling Algorithms

• Coordinator Elections

• Orphan Processes

Page 2: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 2

Distributed Scheduling Algorithm Choices

• Level of scheduling– local scheduling– global scheduling

• Load distribution goals– load balancing– load sharing

• ? Study fig. 7.2 p. 153 ?

Page 3: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 3

Scheduling Efficiency Goals

• Efficiency metrics:– time, execution cost, resource utilization

• Optimal scheduling is NP-Hard.

• Sub-optimal scheduling– sub-optimal approximate solutions – sub-optimal heuristic solutions

Page 4: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 4

Processor Binding Time

• Processor binding time – determines at what point the scheduling algorithm

decides when and where a process will execute.

• Static binding– processor assignment is done once at the link time

• Dynamic binding– process image is relocatable

Page 5: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 5

Scheduling Algorithm Approaches

• Usage points– used with centralized server– usage table on the server contains an entry for

each computer used in the system– usage points are either charged or credited to a

processor• charged if a processor requests utilization of remote

resources

• credited if a processor makes itself available to others

Page 6: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 6

Graph Theory

• Relies on obtaining the minimum cutset for a vertex of a graph.

• ?See fig. 7.5 p.161. How is the processor assignment created?

• Basis for evaluating performance:– minimize total execution and communication

cost– minimize total interference costs.

Page 7: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 7

Probes

• Messages are send to members of a system to locate an appropriate processor to schedule a process.

– Distributed approach– optimal or suboptimal

Page 8: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 8

Scheduling Queues

• Local and global scheduling queues.

• Priority based

• Hints from the user.

Page 9: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 9

Stochastic Learning• Stochastic learning is a heuristic that attempts

to find the best solution based on previous actions (learning from experience)

• Each system state is represented by an automaton vector using workload indicators such as:– one-minute workload averages– amount of free memory– CPU idle time– Length of ready queue

Page 10: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 10

Coordinator Election

• Used when there is a need for an elected centralized server in a distributed system.

• Study box 7.3 and fig 7.7 and explain the Bully algorithm

Page 11: Distributed Process Management1 Learning Objectives Distributed Scheduling Algorithms Coordinator Elections Orphan Processes

Distributed Process Management 11

Orphan Processes

• Orphan process is a child process of a terminated process.

• Exacerbated in D.S. because of RPCs.

• Cleanup of orphan processes:– family trees (study fig 7.8 and box 7.4 p 171)– child process allowance (study fig.7.9 p. 173)