2009 International Conference on Advances in Recent Technologies in Communication and Computing
Design and Implementation of an efficient Two-level Scheduler for Cloud Computing Environment Dr. Sudha Sadhasivam
R. Jayarani
Dept of Computer Science and Engineering, PSG College of Technology, Coimbatore, India e-mail:
[email protected]
Dept of Computer Technology & Applications, Coimbatore Institute of Technology, Coimbatore, India e-mail:
[email protected]
Dr. N.Nagaveni
R. Vasanth Ram,
Dept of Mathematics, Coimbatore Institute of Technology, Coimbatore, India e-mail:
[email protected]
Dept of Electronics and Communication Engineering, PSG College of Technology, Coimbatore, India e-mail:
[email protected]
Abstract—Cloud computing focuses on delivery of reliable, fault-tolerant and scalable infrastructure for hosting Internet based application services. This paper presents the implementation of an efficient Quality of Service (QoS) based Meta-Scheduler and Backfill strategy based light weight Virtual Machine Scheduler for dispatching jobs. The user centric meta-scheduler deals with selection of proper resources to execute high level jobs. The system centric Virtual Machine (VM) scheduler optimally dispatches the jobs to processors for better resource utilization. We also present our thoughts on scheduling heuristics that can be incorporated at data center level for selecting ideal host for VM creation. The implementation can be further extended at the host level, using Inter VM scheduler for adaptive load balancing in cloud environment.
II.
Keywords - Distributed Computing, Cloud Computing, MetaScheduler, Local Scheduler, Conservative Backfilling, Inter Virtual Machine Scheduler.
I.
INTRODUCTION TO CLOUD COMPUTING
Cloud computing delivers infrastructure, platform, and software as services. These services are made available on a subscription basis using pay-as-you-use model to customers, regardless of their location. Cloud computing is a costeffective model for provisioning services and it makes IT management easier and more responsive to the changing needs of the business[1]. The access to the infrastructure incurs payments in real currency in cloud environment. The simulation based approaches provide significant benefits, as it allows researchers to test their proposed algorithms and protocols in a repeatable and controlled environment free of cost, and to find solution to the performance bottlenecks before deploying in the real cloud [2]. The CloudSim toolkit supports First Come First Serve (FCFS) and Round Robin (RR) scheduling strategies for internal scheduling of jobs. FCFS and RR suffer from Long average waiting time for longer jobs which necessitates for the deployment of a better scheduling strategy at the cluster level. 978-0-7695-3845-7/09 $26.00 $25.00 © 2009 IEEE DOI 10.1109/ARTCom.2009.148
BACKFILLING SCHEDULING POLICIES
General concept of backfilling allows smaller jobs to move forward in the schedule as long as such movement does not cause any other scheduled jobs to be further delayed. This section discusses some of the variants of the Backfilling scheduling strategies that can be used at the cluster level. In EASY (Extensible Argonne scheduling sYstem) backfilling, only the first queued job is given Earliest Start Time. Now it is possible to schedule and dispatch the smaller jobs if they would not delay the start of the job in the head of the waiting queue. In the second approach namely, Conservative Backfilling every queued job is given guaranteed start time, so that it has a bounded delay. The third approach namely, Slack based backfilling differs from conservative method by supporting priorities. It assigns each waiting job some slack, which measures the maximal amount of time that the job may be delayed beyond its initially assigned start time. When a job is delayed or sped up its slack changes accordingly. This way the scheduler enjoys more flexibility than conservative scheduling, but still retains the execution guarantee [6]. The conservative backfilling achieves the same result as the slack based method, but it is comparatively light weight. Hence our proposed work implements conservative backfilling at the cluster level for better throughput. III.
DESIGN AND IMPLEMENTATION OF TWO LEVEL SCHEDULER
Our work concentrates on the design of a system that analyzes scheduling techniques for planning various types of jobs in cloud environment. The activities involved in job scheduling for cloud environment includes the selection of processing resource like data center, host and virtual machine and the processing order of jobs for every resource. Some of the constraints to be considered for scheduling include the QoS specifications like deadline, budget, and software licenses of jobs, job dependencies and resource 884
D. Space Shared VMScheduler: Existing Features: The existing local scheduler in cloudsim employs simple First Come First Serve (FCFS) Policy. It is associated with each VM, which places the newly arrived cloudlets in a queue, in case of non availability of required resources. When resources become free, only newly arrived cloudlets are served, but not the queued ones, hence they suffer with starvation. Enhancements: Intra VMScheduler: The above mentioned issue is very well handled in our enhanced Intra VMScheduler, by implementing a typical queue-based policy known as Conservative Backfilling. The proposed Intra VMScheduler, implementing conservative backfilling strategy handles three different cases namely, regular dispatch, backfill and backlog.
limitations. The proposed two- level scheduler focuses on optimizing the system throughput by maximizing the overall resource utilization and guaranteeing increased performance of the applications. The proposed approach extends the cloudsim toolkit [2], by implementing a novel high-level meta-scheduler. As meta-scheduler cannot have a control over the resources at a data center and the full set of jobs submitted to the resources, we implemented a low-level local scheduler to perform efficient job scheduling in cloud environment. Some of the enhancements proposed by our two-level scheduler are discussed in the following paragraphs. A. Datacenter Broker: Existing Features: In Cloudsim, this component randomly selects the datacenter irrespective of their heterogeneity in hardware, software configuration and pricing schemes for usage. Then the broker maps the cloudlets to all the created virtual machines in a circular fashion without considering the Processing Elements (PEs) required by the cloudlets. Enhancements: We have proposed a meta-scheduler that selects the data center, based on user defined QoS specifications such as deadline and budget.
IV.
EXPERIMENTAL SETUP
In this section, we describe the experimental set up for the proposed two-level scheduler in the cloud environment. A.
Cloudsim toolkit CloudSim is a new generalized and extensible simulation framework. It helps in modeling, simulation and experimentation of Cloud computing infrastructure and management services. Cloud Infrastructures such as datacenters, service brokers, and scheduling and allocation policies can be modeled using this toolkit
B. VMProvisioner: Existing Features: The simple VM Provisioner of the CloudSim chooses the host with less PEs in use, as the host for VM. This heuristics ensures load balancing. Nevertheless, many VM Create Requests fail, even though the required numbers of free PEs are available across various hosts. Enhancements: The optimal VM Provisioner in the proposed system rectifies the said problem by optimally creating VMs in the hosts by ordering the request appropriately. The VM creation requests with more resources are allocated followed by the requests with fewer resources, thus minimizing the number of failures in VM creation.
B. Job Categorization criteria The following table explains the typical nature of the arriving jobs in a real–time scenario. There are four different types of job-mix namely, short-narrow (SN), short-wide (SW), long-narrow (LN), and long-wide (LW). TABLE I.
C. Inter VM Scheduler: Existing Features: The host in a data center generally manages the life cycle of VM. It creates the VMs with required configuration before the cloudlets are dispatched and destroys all VMs only when all VMs finish processing the cloudlets. The VM Scheduler in CloudSim does not guarantee load balancing and optimal resource utilization of PEs. In this case a situation may arise wherein a specific VM can be Idle, and other VMs may be heavily loaded with cloudlet execution. Enhancements: In our proposed system, the idle VM can be destroyed and a new VM with required configuration can be created in order to balance the load thereby utilizing all the PEs of a host. This proposed inter VM scheduler takes the role of VM load equalizer in an adaptive manner.
V.
JOB CATEGORIZATION CRITERIA
EXPERIMENTAL RESULTS
The proposed Intra VMScheduler is tested with cloudlets having different Job Length, required PEs and arrival time. The results are compared with existing Space Shared scheduler. The simulated results of comparison are shown in tables 2, 3 and 4 for the different job mixes such as SN, SW, and LN respectively. Our experimental results show that the job mix such as short narrow and short wide are made regular dispatch and long narrow jobs are backfilled and long wide jobs are backlogged predominantly. The proposed scheduler proposed scheduler outperforms the existing space shared scheduler.
885
COMPARISON OF SPACE SHARED SCHEDULER WITH INTRA VMSCHEDULER FOR SHORT NARROW JOBS
Long Narrow Jobs Processing Time (in sec)
TABLE II.
40000
Space Shared VM Scheduler
20000 0 0
1
2
3
Intra VM Scheduler
Cloudlet ID
Figure 3. Comparison of Space Shared VM Scheduler with Intra VM Scheduler for Long Narrow jobs
Processing Time (in sec)
Short Narrow jobs
Space Shared VM Scheduler
300 200
VI.
This paper enhances the existing scheduling strategy in the cloud environment by proposing a two-level scheduler optimizing the performance both in the higher and cluster level. In addition, the novel Inter VM Scheduler, for load balancing at the VM level and optimized VM Provisioner for enhanced resource utilization have been proposed. Experimental results demonstrate that conservative backfilling is highly suitable at the cluster level in grid as well as cloud computing.
Intra VM Scheduler
100 0 0
1 2 Cloudlet ID
3
Figure 1. Comparison of Space Shared VM Scheduler with Intra VM Scheduler TABLE III.
COMPARISON OF SPACE SHARED SCHEDULER WITH INTRA VMSCHEDULER FOR SHORT WIDE JOBS
VII. [1]
[2]
[3]
Processing Time (in sec)
Short Wide jobs Space Shared VM Scheduler Intra VM Scheduler
600 400 200 0
0
1 2 Cloudlet ID
[4] [5]
3
[6]
Figure 2. Comparison of Space Shared VM Scheduler with Intra VM Scheduler for Short Wide jobs TABLE IV.
CONCLUSION AND FUTURE WORK
COMPARISON OF SPACE SHARED SCHEDULER WITH INTRA VM SCHEDULER FOR LONG NARROW JOBS
[7]
886
REFERENCES
R Buyya, C. S. Yeo and Venugopal, “Market oriented cloud computing: Vision, hype, and reality for delivering IT services as computing utilities”. Proceeding of the 10th IEEE International Conference on the High Performance Computing and Communications, 2008. Rodrigo N. Calheiros, Rajiv Ranjan, Cesar A. F. De Rose, and Rajkumar Buyya, “CloudSim: A Novel Framework for modeling and Simulation of Cloud Computing Infrastuctures and Services”, 2009. Rajkumar Buyya, Chee Shin Yeo, Srikumar Venugopal, James Roberg, and Ivona Brandic, “Cloud Computing and Emerging IT Platforms: vision, Hype, and Reality for Delivering Computing as the 5th utility,” 2008. IBM Red Book, ‘Multitenant Utility Computing on IBM Power Systems Running AIX’, 2009. A.W Mualem and D.G Feitelson, “Utilization predictability, workloads and user runtime estimates in scheduling the IBM SP2 with backfilling”, IEEE Trans. Parallel and distributed System 12(6), June 2001. David Talby, Dror G Feitelson, “Supporting Priorities and Improving Utilization of the IBM SP Scheduler Using Slack-Based BackFilling”, IEEE Transactions on Parallel and distributed Systems vol 18, no10 Dec 1999. F. Howell and R Mcnab, “SimJava: A discrete event simulation library for java”. Proceeding of the first International Conference on Web-Based Modeling and Simulation, 1998.
