Distributed Process Scheduling

1
Distributed Process Scheduling

• Sreekanth Padidala

2
Topics.

• A System Performance Model
• Static Process Scheduling
• Dynamic Load Sharing and Balancing
• Real-Time Scheduling

3
Objective.

• The primary objective of scheduling is to enhance
  overall system performance metrics such as
  process completion time and processor
  utilization.
• The existence of multiple processing nodes in
  distributed systems present a challenging problem
  for scheduling processes onto processors and vice
  versa.

4
System Performance model

• Partitioning a task into multiple processes for
  execution can result in a speedup of the total
  task completion time. The speedup factor S is a
  function
• S F(Algorithm System Schedule)
• The unified speedup model integrates three major
  Components
• Algorithm development
• System architecture
• Scheduling policy
• with the objective of minimizing the total
  completion time (makespan) of a set of
  interacting processes

5
System Performance model

• If processes are not constrained by
  precedence relations and are free to be
  redistributed or moved around among processors in
  the system, performance can be further improved
  by sharing the workload
• statically - load sharing
• dynamically - load balancing

6
Static Process Scheduling

• Scheduling a set of partially ordered tasks on a
  nonpreemtive multiprocessor system of identical
  processors to minimize the overall finishing
  time (makespan)
• Except for some very restricted cases scheduling
  to optimize makespan is NP-complete
• Most research is oriented toward using
  approximate or heuristic methods to obtain a near
  optimal solution to the problem
• A good heuristic distributed scheduling algorithm
  is one that can best balance and overlap
  computation and communication

7
Precedence Process Model

• Program is represented by a directed acyclic
  graph (DAG)
• Computational model--this model is used to
  describe scheduling for program which consists
  of several sub-tasks. The schedulable unit is
  sub-tasks.
• Primary objective of task scheduling is to
  achieve maximal concurrency for task execution
  within a program.
• Scheduling goal minimize the makespan time.

8
Precedence Process Model

• Algorithms
• List Scheduling (LS) Communication overhead is
  not considered. Using a simple greedy heuristic
  No processor remains idle if there are some tasks
  available that it could process.
• Extended List Scheduling (ELS) the actual
  scheduling results of LS with communication
  consideration.
• Earliest Task First scheduling (ETF) the
  earliest schedulable task (with communication
  delay considered) is scheduled first.
• what is the scheduling results of the above
  example when there are two processors? how about
  four processors?

9
Communicating Process Model

• There are no precedence constrains among
  processes
• Modeled by a undirected graph G, node represent
  processes and weight on the edge is the amount of
  communication messages between two connected
  processes.
• Scheduling goal maximize the resource
  utilization.

10
Dynamic Load Sharing and Balancing

• No prior knowledge is assumed
• Scheduling need to be dynamic
• Assignment decision made locally
• Based on disjoint process model
• Performance goal for scheduling is the high
  utilization of the system and equal fairness of
  user processes
• Solutions without a centralized controller
  sender- and receiver-initiated algorithms.

11
Sender-Initiated Algorithm

• Activated by a sender process that wishes to
  offload some of its computation
• Transfer policy When does a node become the
  sender
• queue size, ...
• Selection policy How does a sender choose a
  process for transfer
• Newly-arrived processes,...
• Location policy which node should be the target
  receiver?
• Randomly-chosen, polling for
  minimal load,...
• performs well on a lightly loaded system

12
Receiver-Initiated Algorithms

• A receiver can pull a process from others into
  its site for execution
• Activates the pull operation when its
  queue length is shorter than RT
• Require preemption capability
• Which process to remove is not obvious
• Can be effective at highly-loaded systems
• Can be combined with sender-initiated algo.

13
Real-Time Scheduling

• Software hardware for real systems which have
  real-time constraints and are interrupted often.
• Hard Real-time System
• Guaranteed, deterministic behavior.
• Critical Jobs.
• Eg. Nuclear Power Plant controller.
• Soft Real-time System
• High Throughput
• Concurrent access, Large demand.
• Eg. Airline Reservation System.

14
Real-time Scheduling Algorithms

• Classification
• Static vs Dynamic
• Preemptive vs Non-preemptive
• Global vs Local.
• Static
• - Fixed-priority Rate-monotonic algorithm
• - Fixed-priority Deadline-monotonic algorithm
• - Graph based Approach
• Dynamic
• - Earliest Deadline First
• - Least Laxity

15
Research(1995)...

• The problem of process distribution in a computer
  network can be formulated as a distributed search
  process.
• The goal is to find those processors which can
  execute the processes in the most cost-efficient
  way.
• Present a modified version of the AOalgorithm
  using statistical data as a heuristic function.
  Based on previous observations of processors'
  efficiency in process execution the search will
  focus on the most promising search path to locate
  appropriate processors.(Gyires, T 1995)

16
Research(1995)...

• A new process scheduling queue system called the
  distributed queue tree (DQT) for a distributed
  memory, dynamically partitionable parallel
  machines was proposed.
• The combination of dynamically nested
  partitioning and time-sharing scheduling may
  provide an interactive environment and higher
  processor utilization.
• The key idea of DQT is to distribute process
  scheduling (Hori, A.1995) queues to each
  partition.
• A round-robin scheduling algorithm and several
  task allocation policies on DQT was proposed.

17
Research(2003)

• an approach to real-time scheduling of processes
  onto a pool of general-purpose processors.
• The processes are of different criticality
  levels they have different start times,
  execution times and deadlines.
• The proposed(Santiprabhob,P2003) scheduling
  method takes into account processes criticality
  level, processes urgency and load at each
  criticality level when making a dispatching
  decision by means of a fuzzy rule-based system.

18
Research(2003)

• A new generation scheduling paradigm- the
  Internet scheduling environment.
• It is formed by a group of Internet scheduling
  agents which share computational resources to
  solve scheduling problems in a distributed and
  collaborative manner.
• To coordinate computational resource
  collaboration among agents, introducing the
  market-based control mechanism in which
  self-interested agents initiate or participate in
  auctions to sell or buy scheduling problems.
• The experiments (Yen, B.P.-C.,2003) on the
  LekiNET testbed demonstrate that the agent-based
  market-driven Internet scheduling environment is
  feasible and advantageous to future scheduling
  research and development.

19
Research(2006)..

• Process planning and scheduling are considered as
  two separate and distinct phases in
  manufacturing. Traditional approaches to these
  problems do not consider the constraints of both
  domains simultaneously and can only result in
  sub-optimal solutions.
• Most process planning and scheduling systems are
  off-line and centralized. The process plans
  generated off-line may become invalid at the time
  of plan execution. On the other hand, scheduling
  based on rigid process plans may have already
  lost the optimal options.

20
Research(2006)..

• Today, there is an increasing need for the
  integration of process planning and scheduling to
  generate more realistic and practical solutions.
• This paper(Lihui Wang,2006) provides a literature
  review on the integration of process planning and
  scheduling, particularly on agent-based
  approaches and overview of their approach towards
  distributed process planning and scheduling.

21

• Systems, Man and Cybernetics, 1995. 'Intelligent
  Systems for the 21st Century'., IEEE
  International Conference A distributed process
  scheduling algorithm based on statistical
  heuristic search byGyires, T.
• System Sciences, 1995. Proceedings of the
  Twenty-Eighth Hawaii International Conference  
  A scalable time-sharing scheduling for
  partitionable distributed memory parallel
  machines by Hori, A. Maeda, M. Ishikawa, Y.
  Tomokiyo, T. Konaka, H.
• Aerospace Conference, 2003. Proceedings. 2003
  IEEE Fuzzy rule-based process scheduling method
  for critical distributed computing environment by
  Santiprabhob, P. Thumthawatworn, T.
• Systems, Man and Cybernetics, Part A, 2003 IEEE
  Transactions on Internet scheduling environment
  with market-driven agents by Yen, B.P.-C. Wu,
  O.Q.
• International Journal of Computer Applications in
  Technology 2006 - Vol. 26, No.1/2  pp. 3 14 An
  overview of distributed process planning and its
  integration with scheduling by Lihui Wang,
  Weiming Shen, Qi Hao.
• http//en.wikipedia.org/wiki/Real-time_operating_s
  ystem.
• http//www.cdot.ch/thomas/DistOS/lecture6.pdf