Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Copy of Cloud Computing

No description
by

Aboozar Rajabi

on 14 September 2013

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Copy of Cloud Computing

By:
Supervisor:
Nasser Yazdani
ICT industry generates about 2% of total global CO2 emissions.
The energy-related costs amount represents 42% of total data center budget.
About 40% of total energy of a data center is consumed by IT equipment. One-third is related to the network.
Scheduling on Distributed Cloud Data Centers
Placement of VMs in Distributed Data Centers
Introduction
Automatic Resource Provisioning in
Green

Federation of
Clouds

Problem
Related Works
Models
Solution Approach
Data center
Federation
Evaluation
Garg et. al. (2011) have proposed some heuristic solutions.
heuristics consider CO2 emission and profit.
Cloud-oriented data centers are homogeneous.
Network-awareness is ignored.
Some constraints are ignored to make it easier.
-
Placement of VMs in a data center is not considered.
Profit is ignored.
Constraints are limitted.
-
Some important constraints are ignored.
The system is homogeneous.
Network-awareness is not considered.
-
A pareto-based Metaheuristic for Scheduling
Khosravi et. al (2013) have proposed a solution for placement of VMs based on the carbon emission and energy consumption.
Kessaci et. al (2012) have considered geographycally distributed data centers.
Genetic Algorithm (GA) is utilized to solve the optimization problem.
Deadline is the main constraint.
A set "C" of c data centers
Data center parameters:
COP (Coefficient of Performance)
Execution price
CO2 emission rate
Electricity price
Number of servers
A set "P" of p heterogeneous servers
Servers are DVS-enabled
Memory and disk storages are also different
Constraints:
Minimizing energy consumption can significantly reduce the amount of energy bills and then increase the providers' profit.
High energy consumption leads to high carbon emission which means destructive environmental impacts.
A proper scheduling approach is needed for a cloud provider who own multiple data centers.
A simple Task Graph (DAG)
Computation cost of the tasks
Motivation
NIST definition of Cloud Computing:
"Cloud computing is a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned and released with minimal management effort or service provider interaction"
Federation
Aboozar Rajabi
Publications
A set of cloud data centers geographycally distributed around the world.
Service
A set "S" of s services
Each service has a deadline
Each service is a set "N" of n precedence-constrained tasks
Constraints
Memory
Disk Storage
Deadline
No service Redundancy
Energy of federation
Computing equipment
Cooling equipment
CO2 Emission
CO2 Emission of a service
Total CO2 Emissions of the federation
Profit
Profit of a service
Total profit of a cloud provider
Energy of Data center
Power Consumption of a Server
Power consumption of Network
The energy consumption of network is dominant by switches.
"Decision Support-as-a-Service: An Energy-aware DSS in Cloud Computing",
5th International Conference on Information and Knowledge Technology (IKT 2013).
"Communication-aware and Energy-efficient Resource Provisioning for Real-Time Cloud Services", The 17th CSI International Symposium on Computer Architecture & Digital Systems (CADS 2013).
Invited to extend for JCSE journal
"Towards Energy-aware Resource Scheduling to Maximize Reliability in Cloud Computing Systems", 15th IEEE International Conference on High Performance Computing and Communications (HPCC 2013).
"EAICA: An Energy-aware Resource Provisioning Algorithm for Real-Time Cloud Services", 6th IEEE CloudCom (CloudCom 2013), Submitted.
Architecture
Proposed architecture consists of two schedulers at two different levels.
ICA + MELP heuristic
ICA (Imperialist Competitive Algorithm) is a socio-politically inspired optimization strategy which starts by an initial population.
Popoulation individuals called country are divided into two groups: colonies and imperialists.
Federation-level
Multi-objective optimization problem
Objectives:
Minimizing CO2 emissions
Maximizing Provider's profit
Scheduling period: 60s
ICA is executed with the defined cost function
ILP (Federation-level)
Minimize TCO2E(X)
and
Maximize TProf(X)
Subject to (Memory), (Disk) and (Deadline)
Data center-level
ILP (Data center-level)
Minimize TE(X)
Subject to (Memory), (Disk), (Precedence) and (Deadline)
CCR (Communication to Computation Ratio)
Indicates whether a task graph is communication-intensive, computation-intensive or moderate.
Computed by the average communication cost divided by the average computation cost.
Low CCR Sort
Computation cost is dominant.
When the CCR <= 1, servers are selected from all over the data center.
Sudo code:
1. Compute the "Server Score".



2. Sort by the "Server Score" in increasing order.
3. Select the servers as many as needed from the sorted list.
High CCR Sort
Communication cost is dominant.
When the CCR > 1, servers are selected from a rack, then from other racks of a same cluster and finally from other clusters.
Sudo code:
1. Compute the "Server Score".
2. Sort servers of each rack by the "Server Score" in increasing order.
3. Select from the first rack, then the second and ... .
First Phase:
Consolidation
The energy cost function is ignored to find the required number of servers.
Until the constraints is met, the ICA will be executed iteratively.
Number of servers will be increased by one in each iteration while it reaches the required number.
Second Phase:
DVS
ICA considers the energy as well as other constraints.
ICA is run for number of required servers found in the first phase.
The DVS is utilized when it is feasible.
The feasibility depends on the deadlines.
Configuration
Data Center Scheduler
Federation Scheduler
Federation
Data Center
A data center consists of some heterogeneous DVS-enabled servers.
Service
Services are selected from Feitelson's Parallel Workload Archieve (PWA)
Submission Time and Run Time is extracted.
4 classes of service arrival rates is defined:
Low (X)
Medium (10X)
High (100X)
Very High (1000X)
3 forms of DAGs are utlized as benchmarks:
Laplace Equation Solver (LES)
Gaussian Elimination (GE)
Fast Furier Transformation (FFT)
Alpha and Beta coefficients are used to weight the objectives.
Alpha = 1: CO2 emission is more important.
Beta = 1: Profit is more important.

1. Evaluation of Low & High CCR Sorts
2. Comparisons
3. Proposed architecture for the scheduler
Conclusion
The problem of scheduling real-time services in distributed cloud data centers is investigated.
The addressed services presents precedence constrained parallel applications.
Two schedulers which are located at two levels are proposed as the architecture of solution.
Each schedulers utilizes ICA and solve an optimization problem with different objectives.
Simulation results show about 9% improvement in comparison with related works.
Memory:


Disk storage:

Deadline:


Precedence:
No task redundancy
Avg. Energy Saving of the ICA in comparison with:
GA: 9.10%
Pure DVS: 3.61%
Pure Consolidation: 17.74%
NPA: 82.46%
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Most Execution Time-Least Power (MELP) Heuristic
Sudo code
Scheduling tasks with higher execution times on servers with lower power consumption leads to energy saving.
MELP heuristic sorts servers and tasks by their maximum power and average execution times respectively.
Scheduling will be done based on these two lists.
Solution Representation
Full transcript