Scheduing Algorithms

1
Scheduing Algorithms

• Vadik Vargas

2
Introduction

• Purpose
• To show through algorithms examples that there is
  no best algorithm, but that different
  algorithms work best for different applications.

3
Scheduling Algorithms Overview

• Temporary Data Deadline-Based Predictive
  Algorithm (TDDBPA)
• Greedy Fair Queuing (GrFQ)
• Scheduling Tree
• Power-Aware Speculative Algorithms with AND/OR
  Graphs (Static and Adaptive).
• General Bipartite Graphs Algorithm GGP and
  Optimized General Bipartite Graphs Algorithm
  OGGP.

4
Temporary Data Deadline-Based Predictive
Algorithm (TDDBPA)

• Introduces Temporary Deadlines
• Processes that meet these deadlines are run,
  otherwise they are cancelled or postponed
• Temp. Deadlines depend on how fast the data is
  refreshed and if it is in condition to be read
  or written flags to signal data state.
• Perfect for applications where data is very
  sensitive such as sensors Navigational systems,
  fast decision making systems (defense).

5
Greedy Fair Queuing (GrFQ)

• Emphasizes fair allocation of available
  bandwidth.
• Examines and decides what process to run in a
  per-packet basis, achieving a better fairness
  with lower computational complexity.

6
Scheduling Tree

• Designed to allow processes access to the
  processor for determined time slots (quanta)
• Implemented as a hierarchical composition of RR
  in a tree, tries to meet each process time
  demands.
• Communication systems may need some of the
  expensive processes to go to sleep most of the
  time when not being used.

7
Power-Aware Speculative Algorithms (Static and
Adaptive)

• Try to save power by limiting the number of
  voltage switches necessary to run a process
• Calculates best speed to run process. Uses
  statistics, and speed may be set to change
  half-way into the process (Static), or new speeds
  may be calculated while the process is running
  (Adaptive).
• Good were power consumption must be limited. Ex
  running a machine in auxiliary power.

8
General Bipartite Graphs Algorithm GGP and
Optimized General Bipartite Graphs Algorithm OGGP

• Designed to schedule processes in multiprocessor
  systems that run processes at the same time
  (parallel).
• Addresses the issue of scheduling messages
  efficiently when trying to redistribute data
  between two clusters.
• Detailed discussion

9
Bipartite Graphs Algorithms Discussion

• Looks at architecture of the problem as composed
  of clusters (groups of processors/terminals)
  interconnected by a backbone (high speed network)

10
Problems to solve

• Parallelism problem
• Given a backbone with a determined capacity for
  traffic.
• Sum of transmission size from nodes in a cluster
  is too big for the backbone to handle.
• Most systems let the backbones transport layer
  try to handle the excessive traffic (TCP), but
  usually results in lost data, or causing the
  transmission window size to be reduced wasting
  transmission capacity.

11
Characteristics of Solution

• Low complexity easy to implement
• Communication between clusters is done by node
  pairs (step)
• Total communication size must not be bigger than
  the backbones capacity.
• Adding preemption
• Making minimum quantum as small as the smallest
  step transmission.
• Organizing transmission according to Bipartite
  Graphs guarantees that total number of steps use
  maximum backbone capacity, but not more.

12
Bipartite Graph Example