application scheduling in cloud sim
Post on 05-Dec-2014
2.917 Views
Preview:
DESCRIPTION
TRANSCRIPT
1
Application Scheduling in CloudSim
Presented by: Pradeeban Kathiravelu
Supervised by: Prof. Luís Veiga
Implementation of Distributed Systems
2
Application Scheduling
Scheduling an application– to be executed– using a resource– in a cloud environment
3
Aim
Evaluating the Scheduling algorithms– Strict matchmaking-based– Utility-driven
4
Aim
Evaluating the Scheduling algorithms– Strict matchmaking-based– Utility-driven
Criteria– Mean execution time– Mean user submission time– Average resource utilization– Job Scheduling Success Ratio
5
Objective Function Algorithm→
Strict matchmaking-based– Minimum Execution Time (MET)– Minimum Completion Time (MCT)– Maximum Resource Utilization
– Matchmaking– First-come first-served (FCFS)– Round Robin (RR)
6
Utility Algorithm→
User Satisfaction Partial Requirement Satisfaction.
– Number of metrics– Are they equally important?
7
Evaluation
CloudSim– Simulation tool for cloud
computing Representing by objects.
8
CloudSim
Cloudlets– The applications/tasks
Processing Elements (Pe:s)– The CPU
Hosts Virtual Machines Datacenters
– Infrastructure Provider
9
DatacenterBroker
10
Experiments
2 → 200 users 2 data centers
– 2 hosts each– OS, Arch, VMM
5 → 20 VMs– 200 → 1000 MIPS
20 → 40,000 Cloudlets– With varying lengths– 100 → 4000 MI
11
E1: VM and Host Level Scheduling
200 users 5 VMs
– 200, 400, 600, 800, 1000 MIPS 4000 Cloudlets
– 100 → 4000 MI Change the VM and Host level
scheduling. {FCFS, RR}
12
Start Time
13
Finish Time
14
E2: Application Scheduling Algorithms
RR and FCFS– With and without over-subscription
Maximum Resource Utility Dynamic Allocation
– With partial requirement satisfaction
– OS, VM, MCT
15
Completion Time and Execution Time
200 users 5 VMs
– 200, 400, 600, 800, 1000 MIPS 4000 Cloudlets
– 100 → 4000 MI– Varying requirements and utility
No time limitation Maintain 100% Job Success Ratio
16
Mean Submission Time and Mean Execution Time
17
Summary
Each algorithm performs better for– different criteria– different tasks
Utility-driven algorithms with Partial requirement satisfaction take the lead.
18
Summary
Each algorithm performs better for– different criteria– different tasks
Utility-driven algorithms with Partial requirement satisfaction take the lead.
Thank you!
19
Summary
Each algorithm performs better for– different criteria– different tasks
Utility-driven algorithms with Partial requirement satisfaction take the lead.
Thank you!
Questions?
top related