Chapter 11: Distributed Scheduling

Load

• Resource and CPU queue lengths are good indicators of load.
• Artificially increment CPU queue length for transferred jobs on their way.
• Set timeouts for such jobs to safeguard against transfer failures.
• Little correlation between queue length and CPU utilization for interactive jobs: use utilization instead.
• Monitoring CPU utilization is expensive.

Classification of Algorithms

Static

: decisions hard-wired into algorithm using prior knowledge of system.

Dynamic

: use state information to make
decisions.

Adaptive

: special case of dynamic algorithms; dynamically change parameters of the
algorithm.

Need to consider overheads of system state collection.

Load Balancing vs. Sharing

Load Sharing

: reduce the likelihood of
unshared state by transferring tasks to lightly loaded nodes.

Load Balancing

: attempt to equalize load at all computers.

Load balancing incurs higher overheads.




Anticipatory task transfers can reduce the
duration of unshared state.

Preemptive vs. Nonpreemptive

Preemptive transfers

: transfer of a task that is partially executed.

• expensive due to collection of task's state

Nonpreemptive transfers

: only transfer tasks that have not begun execution.

Components of Load Distribution

Transfer policy

: threshold based.

Selection policy

: overhead in transfer of
selected task should be offset by reduction in its response time.

Location policy

: possibly use polling to find suitable node.

Information policy

: demand-driven, or
periodic, or state-change-driven.

Stability

: queueing-theoretic, or algorithmic perspective.

Information Policy

Demand-driven

: nodes gather information about other nodes.

• sender initiated
• receiver initiated
• symmetrically initiated

State-change-driven

: nodes disseminate
information when their state changes.

Sender Initiated Algorithms

• overloaded node: when a new task makes the queue length exceed threshold, T
• underloaded node: if accepting a task will not cause queue to exceed threshold, T
• overloaded node attempts to send task to underloaded node
• only newly arrived tasks considered for
  transfers
• location policies: random, threshold, or
  shortest
• information policy: demand-driven
• render the system unstable at high load

Receiver Initiated Algorithms

• if a task departure drops node queue length below T: receiver
• underloaded node tries to obtain task from overloaded node
• initiate search for sender either on a task departure or after a predetermined period: why?
• information policy: demand-driven
• remain stable at high and low loads
• drawback: most transfers are preemptive and expensive, especially if round-robin CPU scheduling is employed at nodes.

Symmetrically Initiated Algorithms

• senders search for receivers:
  • successful in low load situations
  • high polling overhead in high load situations
• receivers search for senders:
  • useful in high load situations
  • preemptive task transfer facility is necessary