This article will help you understand the class hierarchies available in Cloudsim for Virtual Machine and Cloudlet Scheduling. This blog post is supported with a video lecture provided at the end, to have a better understanding of the implementation. Let’s start with the basic terms.
Basics of Scheduling
In computers, Scheduling is a process of arranging the submitted jobs/task into a very specific sequence of execution. It is an essential characteristic of any software operating environment, which is handled by a very special program known as a scheduler.
Scheduler’s main objective is to keep the underlined hardware resources(primarily processor) to be used effectively as well as efficient. In general, the scheduler may prefer to have any of the following scheduling approaches:
- Space-shared: In this, the requested resources are allocated dedicatedly to the requesting workload for execution and will be released only on completion. Space-shared is also known as a batch process scheduling.
- Time-shared: In this, the requested resources would be shared among more than one workload(task). The sharing is done based on time-sliced allocation where each workload is allocated with a required resource for a defined time(e.g., 200 milliseconds). Once the defined time slice is over, the current workload execution paused, and the resource is released. The released resource gets allocated to the next workload for the same defined time slice, and this cycle goes on till the time all the workloads execution is over. Time-shared is also known as round-robin scheduling.
Scheduling in Cloud
As cloud computing is the virtualized operating environment, and the virtual machines are the primary computing component which is responsible for the execution of the workloads(tasks). The virtual machine(s) are powered by a physical server host machine (i.e.) hardware. Depending on the requirement of the Virtual Machine(VM) there could be ‘one to one’ or ‘many to one’ mapping between the VM and host machine. That means in cloud computing the scheduling is done at both the mapping levels that are:
- Virtual Machine to Host Machines
- Tasks to Virtual Machines
Both of VM to Host as well as Workload(task) to VM mappings may utilize space-share or time-shared or any other specialized scheduling algorithm.
Scheduling in Cloudsim
The Cloudsim simulation toolkit framework has effectively addressed the Scheduling scenario and implemented it as a set of the programmable class hierarchies with parent class as:
Also, Virtual Machine(VM) and Task( Cloudlet) scheduling are one of the most important and the popular use case to be simulated by researchers using the CloudSim simulation toolkit.
Note: In cloudsim, the task is called as cloudlet, therefore in the following text instead of ‘task’ we will be using the ‘cloudlet’.
Cloudsim Virtual Machine Scheduling
The VmScheduler is an abstract class that defines and implements the policy used to share processing power among virtual machines running on a specified host. The hierarchy of the cloudsim virtual machine scheduler classes is as:
These classes can be located in “org.cloudbus.cloudsim” package of cloudsim. The definition of this abstract class is extended to the following types of policies implemented as classes:
- VmSchedulerTimeShared: This class implements the VM scheduling policy that allocates one or more processing elements to a single Virtual machine and allows the sharing of processing elements by multiple virtual machines with a specified time slice. This class also considers the overhead of VM allocation switching(similar to context switching) in policy definition. Here, the VM allocation will fail if the number of processing elements requested is not available. for example, if the VM request for quad-core processor whereas the allocated host has an only dual-core the allocation will fail.
- VmSchedulerSpaceShared: This class implements the VM scheduling policy that allocates one or more processing elements to a single virtual machine, but this policy implementation does not support sharing of processing elements (i.e.) all the requested resources will be used by the allocated VM till the time the VM is not destroyed. Also, Under this allocation policy, if any virtual machine requests a processing element and is not available at that time, the allocation fails.
- VmSchedulerTimeSharedOverSubscription: This is an extended implementation of VMSchedulerTimeShared VM scheduling policy, which allows over-subscription of processing elements by the virtual machine(s) (i.e.) the scheduler still allows the allocation of VMs that require more CPU capacity that is available. And this oversubscription results in performance degradation.
The application of the VmScheduler classes is while instantiating the host model. Following is the code snippet used in CloudsimExample1.java from line number 160 to 174:
int hostId = 0; int ram = 2048; // host memory (MB) long storage = 1000000; // host storage int bw = 10000; hostList.add( new Host( hostId, new RamProvisionerSimple(ram), new BwProvisionerSimple(bw), storage, peList, new VmSchedulerTimeShared(peList) ) ); // This is our machine
This is where the processing element list is passed as a parameter to the VmSchedulerTimeShared() class call and during the simulation, the cloudsim will simulate the timeshare behavior for the virtual machines. Also, in case you want to test other VmScheduler you may replace it with VmSchedulerTimeShared() call with appropriate parameters, this includes your own designed custom virtual machine scheduler.
Cloudsim Cloudlet Scheduling
The “CloudletScheduler” is an abstract class that defines the basic skeleton to implement the policy to be used for cloudlet scheduling to be performed by a virtual machine. The hierarchy of the cloudsim Cloudlet scheduler classes is as:
These classes again exist in “org.cloudbus.cloudsim” package of cloudsim. The definition of this abstract class is extended as the following types of policies implemented as three individual classes in cloudsim:
- CloudletSchedulerSpaceShared: This class implements a policy of scheduling for Virtual machine to execute cloudlet(s) in space shared environment (i.e.) only one cloudlet will be executed on a virtual machine at a time. It means cloudlets share the same queue and requests are processed one at a time per computing core. Space-sharing is similar to batch processing.
- CloudletSchedulerTimeShared: This class implements a policy of cloudlet scheduling for Virtual machines to execute cloudlets in a time-shared environment (i.e.) more than one cloudlet will be submitted to the virtual machine and each will get its specified share of time. It means several requests (cloudlets) are processed at once but they must share the computing power of that virtual machine(by simulating context switching), so they will affect each other’s processing time. It basically influences the completion time of a cloudlet in CloudSim. Time-sharing is probably referring to the concept of sharing executing power (such as CPU, logical processor, GPU) and is commonly known as the round-robin scheduling.
- CloudletSchedulerDynamicWorkload: This implements a special policy of scheduling for virtual machine assuming that there is just one cloudlet which is working as an online service with a different requirement of workload as per the need of peak/offpeak user load at a specified period of time.
The application of the CloudletScheduler classes is while instantiating the Vm model. Following is the code snippet used in CloudsimExample1.java from line number 82 to 91:
int vmid = 0; int mips = 1000; long size = 10000; // image size (MB) int ram = 512; // vm memory (MB) long bw = 1000; int pesNumber = 1; // number of cpus String vmm = "Xen"; // VMM name Vm vm = new Vm(vmid, brokerId, mips, pesNumber, ram, bw, size, vmm, new CloudletSchedulerTimeShared()); // create VM
By instantiating the CloudletSchedulerTimeShared() class, the Virtual machine is decided to follow the timeshare(round-robin) approach while simulation for scheduling & executing the Cloudlets. Also, in case you want to test other CloudletScheduler you may replace it with CloudletSchedulerTimeShared() call with appropriate parameters, this includes your own designed custom cloudlet scheduler.
Now in case you want to implement your own scheduling policies with respect to Virtual Machine or Cloudlet(s), you may simply extend the VmScheduler or CloudletScheduler class to implement all the abstract methods as specified. This gives you the flexibility to design and implement your own set of algorithms and then later test & optimize during the repetitive simulation runs.
In case you missed reading our popular article on Detailed introduction to Cloudsim you may follow CloudSim Simulation Toolkit: An Introduction.
You may subscribe to an online self-paced course named Learn Basics of Cloudsim, The content for this course which will be updated weekly till August 2021.
Hello sir your tutorial is awesome. please make a video to implement a new scheduling algorithm from the scratch. I am very thankful if u make it for us .
thank you for your kind feedback. I have noted your suggestion and will try to accommodate it ASAP.
Hello Sir. Your series of video lectures really helped a lot for those who are working for load balancing in cloud computing. It is also very helpful to understand the very basic of cloud computing.
I guess, like me. every beginners have the similar sort of doubts that how to implement a new scheduling algorithm with per say 4 VMs and each having 4, 2 and 1 processing cores with varying cloudlets. I will grateful to you and we all will be benefited from it.
Thanks for sharing your views. It brings me a great sense of satisfaction that this work in some way in helping out the research community. I have been working on a couple of related topics and hope it will be posted by August 31, 2019.
Your array of videos on superwits academy helped me to get through it thoroughly. I am working on CloudSimExample6. I have to find out Makespan, Utilization, Avg. Utilization, Load of a VM, Capacity of a VM. Where I should modify the code or where should I write the code to calculate all these?
I have to calculate the cosine similarity between Task and Resources. So, can I take the attributes like CPU, RAM, Storage, BW for a task? I have to define all these parameters for each and every task separately. Can it be possible?
Awaiting your reply.
I have project to implement priority task scheduling in cloud computing. I am using Cloudsim but I don’t know when I add this algorithm and what change I do in cloudsim.
Please help me