Meta-heuristic Based Cloud Resource Provisioning Approach Agenda - - PowerPoint PPT Presentation

Meta-heuristic Based Cloud Resource Provisioning Approach

Agenda for Today

• Introduction
• What is meta-heuristics?
• How is it different from heuristics?
• What makes them appropriate for resource provisioning

as compared to heuristics?

• Using a meta-heuristics approaches to solve the resource

provisioning problem

– Problem statement formulation and analysis of its complexity

• What are the research question that are answered by

meta-heuristics algorithms?

• What are the open research questions in this area?

Introduction

• The cloud computing paradigm is increasingly

becoming mainstream and a growing number

• f companies and research organizations seek

to gain value from its unique characteristics, service models, and deployment forms.

• This gives rise to many different optimization

problems both from the consumers’ and providers’ perspective.

Sample Problems

• Cloud service providers need algorithms that can

– reduce operational expenses and improve economies of scale to maximize profits and provide competitive prices in a highly competitive market. – algorithm that can predict for incoming request of VMs from the cloud users; – Algorithms that help improve utilization of the system – Efficient algorithms to optimize the current VM allocation.

Sample Problems

• Cloud service users also need algorithms

– prediction model to react to the dynamic price model of the CSPs – Algorithms that allow them to efficiently negotiate SLA with the Cloud service providers

• Combined optimization issues

– Algorithms for multi-criteria optimizations such as maximizing QoS while at the same time maximizing resource utilization

Agenda for Today

• Introduction
• What is meta-heuristics?
• How is it different from heuristics?
• What makes them appropriate for resource provisioning

as compared to heuristics?

• Using a meta-heuristics approaches to solve the resource

provisioning problem

– Problem statement formulation and analysis of its complexity

• What are the research question that are answered by

meta-heuristics algorithms?

• What are the open research questions in this area?

What is a Meta-heuristic Method?

• Meta-heuristic methods are one of the common

strategies for solving NP-hard optimization problems

• Meta-heuristic  Heuristic + Randomization
• We use the definition given by [Osman and Laporte

1996] – “An iterative generation process which guides a

subordinate heuristic by combining intelligently different concepts for exploring and exploiting the search space, learning strategies are used to structure information in

• rder to find efficiently near-optimal solutions. “

Meta-heuristic Properties

• Meta-heuristics properties

– not problem-specific. – efficiently explore the search space to find (near-)optimal solutions. – approximate and usually non-deterministic.

Meta-heuristic Approaches

• There are many meta-heuristic approaches that

range from simple local search procedures to complex learning processes

– simulated annealing (SA), [e.g., single solution approach] – evolutionary algorithms (EA), [e.g., population-based ] – ant colony optimization (ACO), [eg, of swarm intelligence] – particle swarm optimization(PSO) [e.g., population-based and swarm intelligence approaches]

• Many different new variants are continually being

proposed.

• We will discuss Particle Swarm Optimization

Particle Swarm Optimization

• Particle Swarm Optimization (PSO) [Kennedy and

Eberhart 1995] algorithm

– An adaptive method – Characterized by simplicity and effectiveness in wide range of application with low computational cost. – It also has fewer algorithm parameters than genetic algorithm. – PSO algorithm works well on most global optimal problems.

• The above characteristics make PSO appealing to

solve cloud resource allocation problem.

– has become popular in solving resource allocation

• ptimization problem

Particle Swarm Background

• The particles are randomly placed

in a problem search space.

• A search area specifies an area that

contains all possible solutions of a combinatorial problem.

• Each particle in search space

adjusts its “flying” pattern according to its own flying experience as well as the flying experience of other particles

Particle Attributes

• Each particle will have

– Velocity: The particle flying speed, which directs the flying of the particle. – Fitness: a fitness value (the objective function value), which will be evaluated by a fitness function to be optimised in each iteration. – Particle position (pbest): Each particle knows its best position (this is the best fitness value so far reached by the particle) – Group position (gbest): The best position so far among the entire group of particles (this is the best particle in terms of fitness in an entire population)

Particle Swarm Background

• PSO is based on a set of flying

particles that constitute a swarm move within the search space looking for the best solution.

• Movement towards a

promising area to get the global optimum

Velocity Update

• Each particle adjusts its velocity (ie. travelling speed)

dynamically corresponding to the flying experiences of itself and its neighbors

pbest velocity gbest

• At each time step, a particle moves toward its pBest

and gBest.

Velocity Update..

Velocity and Position Updates

Illustration of PSO Process

From Andry Pinto, et. al

Illustration of PSO Process

From Andry Pinto, et. al

Illustration of PSO Process

From Andry Pinto, et. al

Illustration of PSO Process

From Andry Pinto, et. al

Illustration of PSO Process

From Andry Pinto, et. al

Illustration of PSO Process

From Andry Pinto, et. al

Particle Swarm Optimization Pseudo code

Algorithm: PSO INPUT: N: number of tasks, I: number of iterations OUTPUT: 1. D[]=N //particle dimension 2. Initialize position randomly 3. Initialize velocity randomly 4. REPEAT 5. Evaluate fitness of individual particle. 6. IF fitness is better than pbest 7. pbest = fitness ; 8. END 9. gbest = neighboughr (pbest); 10. Modify velocities based on personal best and global best. 11. Modify position based on current location and velocity. 12. UNTIL Terminate on some condition. 13. Return result 14. END algorithm

Particle Swarm Optimization

Initialization step F(p) F(p) F(p) Update velocity, position, gbest, pbest The set of particles

# of iterations

Convergence

Agenda for Today

• Introduction
• What is meta-heuristics?
• How is it different from heuristics?
• What makes them appropriate for resource provisioning

as compared to heuristics?

• Meta-heuristics Cloud resource provisioning approach

– Problem statement formulation and analysis of its complexity

• What are the research question that are answered by

meta-heuristics algorithms?

• What are the open research questions in this area?

Preliminaries

Problem Formulation

Illustration

1 1 1 1 1 1 1 1 1

• Which task to machine assignment is the best?

NP Hardness

• Even though this problem has been intensively

investigated, exact polynomial algorithms have not been found yet.

• Furthermore, we can verify that the problem is NP-

hard by reduction to the partition set problem, that, even for n = 2, this problem is NP-hard.

• Meta-heuristic optimization algorithm can be used to

solve NP-hard problems.

Meta-heuristic Resource Allocation

• Application of meta-heuristic techniques to Cloud

Resource Allocation

MalaKalraaSarbjeetSingh, A review of metaheuristic scheduling techniques in cloud computing, Egyptian Informatics Journal, Volume 16, Issue 3, November 2015, Pages 275-295

League Championship Algorithm (LCA)

Cloud Resource Allocation

1 1 1 1 1 1 1 1 1

Applying PSO to Resource Provisioning

Particle in PSO 1 2 3 4 5 6 7 8

A representation of task to resource mapping as a PSO particle Task to resource mapping instance

• Each position in the particle represents a task and the value at

that position represents the mapping of the task to a resource.

• Thus the particle represents mapping of resource to a task.

Open Problems

• Research question 1: Develop a hybrid algorithm that

combines metaheuristic techniques and classical Artificial Intelligence (AI) and Operation Research (OR) methods such as greedy algorithm, backtracking techniques, beam search or constrained programming for cloud data center resource management.

• Research question 2: Investigation and development of a

hybrid approach that combines single solution approach (eg., simulated annealing (SA) and population-based metaheuristic technique for cloud data center resource management.

• Research question 3: Metaheuristic algorithm that can (i)

predict for incoming request of VMs from the cloud users; (ii) prediction model to react to the dynamic price model of the CSPs

Open Problems

• Research question 3: Cloud service providers

(CSPs) aim to reduce operational expenses and improve economies of scale to maximize profits and provide competitive prices in a highly competitive market.

– Very little work exists in the space of meta- heuristic algorithm with respect to reducing

• perational expenses and improving economies of

scale to maximize profits and provide competitive prices

Cloud broker

• In the context of this ecosystem, the role of cloud

broker schemes has been intensively discussed, in particular with regard to decision support. That is, a cloud broker (e.g., in form of decision support or a third-party) can interact between consumers and providers as well as between consumers to increase the value creation.

• Research question 4: Metaheuristics for cloud

brokerage specially in Multi-Cloud Computing

Hybrid Meta-heuristics Algorithm for Resource Allocation

• Although POS’s simplicity is appealing, its suffers from

– Tendency to a fast and premature convergence in mid optimum points – Slow convergence in refined search stage (weak local search ability)

• A hybrid algorithm (PSO–GELS) based on the following

paper for our scheduling problem to decreases makespan and minimizes the number of tasks that miss their deadlines.

Zahra Pooranian, Mohammad Shojafar, Jemal H. Abawajy and Ajith Abraham, Journal of Combinatorial Optimization, Volume 30 Issue 3, October 2015, Pages 413-434

Hybrid Meta-heuristics Algorithm for Resource Allocation

• The proposed scheduling algorithm uses PSO as the

main search algorithm, while gravitational emulation local search (GELS) is used to improve the population.

• Reasons for using both algorithms.

– First, we need an algorithm that is based on a population that can search the entire search space for this problem. – Second, the cloud environment is dynamic, so the scheduling algorithm must be fast enough to adapt with the natural cloud environment and must be able to converge faster than other algorithms. – Moreover, although PSO is weak for local searches, our combination of PSO with an algorithm that is strong in

Evaluation

• Default # of particles = 25
• Number of iterations = 20
• On each evaluation, the pbest and gbest

values are updated according to Equation (1) and Equation 2.

• The evaluation is carried out in a loop until the

results converge or until the specified number

• f iterations (user specified stopping criteria).

Makespan average

SA: simulated annealing GA: Genetic algorithm

Number of tasks Number of machines

Execution Time Analysis

Open Problems

• Research question 5: Develop meta-heuristic algorithm for

federated cloud resource provisioning.

• Research question 6: Which meta-heuristic algorithm is

appropriate for scientific workload on cloud computing?

• Research question 7: How do the meta-heuristic algorithm

perform under spot instance provisioning environment?

• Research question 8: In the studied literature, most of the

authors have focused on reduction of makespan and execution cost whereas others have given significance to response time, throughput, flowtime and average resource utilization.

– Develop a SLA and cost-aware meta-heuristic resource provisioning a task scheduling approach that is tailored for Big Data applications in the Cloud.

• Research question 9: An efficient selection

and utilization of cloud providers and cloud services in Federated Cloud environment.

• Research question 10: Application of

metaheuristics in fog computing environments

• Research question 10: Develop reward-based

adaptive resource management for Federated Cloud computing

References

• Zahra Pooranian · Mohammad Shojafar · Jemal H. Abawajy · Ajith Abraham, An

efficient meta-heuristic algorithm for grid computing

• Habib Shah, Tutut Herawan, Rozaida Ghazali, Rashid Naseem, Maslina Abdul Aziz,

Jemal H. Abawajy: An Improved Gbest Guided Artificial Bee Colony (IGGABC) Algorithm for Classification and Prediction Tasks. ICONIP (1) 2014: 559-569

• Nazri Mohd Nawi, Abdullah khan, M.Z. Rehman, Maslina Abdul Aziz, Tutut

Herawan, and Jemal H. Abawajy,” Neural Network Training by Hybrid Accelerated Cuckoo Particle Swarm Optimization Algorithm”, Springer International Publishing Switzerland 2014 .

• Nazri Mohd Nawi, Abdullah Khan, M. Z. Rehman, Maslina Abdul Aziz, Tutut

Herawan, Jemal H. Abawajy: An Accelerated Particle Swarm Optimization Based Levenberg Marquardt Back Propagation Algorithm. 2014: 245-253

• Nazri Mohd Nawi, Abdullah Khan, M. Z. Rehman, Maslina Abdul Aziz, Tutut

Herawan, Jemal H. Abawajy: Neural Network Training by Hybrid Accelerated Cuckoo Particle Swarm Optimization Algorithm. 2014: 237-244

Thank You… Questions, Comments, …?

Practical Today

• This lab, we implement the PSO scheduling
• You can download the code from here and try

it

– https://www.javatips.net/api/CS249_Workflow_pr

• ject-master/cloudsim-

3.0.1/sources/org/cloudbus/cloudsim/network/da tacenter/PSO.java