KR101350755B1 - Cost Based Scheduling Algorithm for Multiple Workflow in Cloud Computing and System of The Same - Google Patents

Abstract

Includes task pools, schedulers, and executors to reduce total processing time and improve average availability for cloud computing environments. Task pools manage multiple workflows for dynamic cloud services and manage QoS requirements from users. When a workflow is submitted, it calculates the expected execution time and cost of all tasks and stores the tasks in a queuesheet, the scheduler calculates the tasks in the queuesheet according to a multi-workflow scheduling strategy, and the executor executes the tasks in the queuesheet. Select the optimal resource or service to run the task, feed back the task completion status to the task pool when the service of the randomly selected task completes successfully, and remove the task when the task completion status is fed back from the executor. Everything from cloud computing It provides a cost-based method for scheduling the workflow.

Description

Cost Based Scheduling Algorithm for Multiple Workflow in Cloud Computing and System of The Same}

The present invention relates to a cost-based scheduling method for multiple workflows in cloud computing, and more particularly, multi-working in cloud computing environment suitable for cloud computing environment because priority is determined by a cost value with throughput using estimated execution time. A cost based scheduling method for a flow and a system thereof.

In general, a cloud computing structure is composed of a user and a cloud structure, and the cloud structure is formed by having a data center.

In the cloud architecture, the cloud has a single access point through the web browser of the web server, and the data center consists of the service resources used by the service provider to service the application for the user.

In the cloud computing architecture, the service requested by the user is transmitted through a web browser, and the requested service arrives at the data center and is processed. It is composed.

Cloud computing is extensively scalable and consists of abstract entities that support different levels of service to users, runs dynamically using virtualization techniques, and provides services to multiple users at the same time. .

For workflow scheduling in cloud computing, first, the cloud must provide the same service to multiple users, so it is necessary to be able to schedule multiple workflows with various quality of service (QoS) parameters. Since there is always the possibility of a flow instance, the scheduler must be able to schedule multiple workflows, and thirdly, users must have access to resources or services whenever they want, so the cloud service is ready to start a new workflow at any time. Consideration should be given to new features such as

How to efficiently allocate resources in cloud computing to get the maximum benefit is the ultimate goal of cloud computing service providers, and cloud computing needs to be limited and appropriately assigned multiple applications of many users at the same time.

However, as more users or applications are used, cloud services degrade stability and performance. For example, on February 15, 2008, Amazon's server and storage computing service, S3, was shut down for about two hours, killing thousands of businesses, developers, and more than 300,000 users. As cloud computing has become more common in recent years, its users and uses have become more widespread, there is a growing concern that the damage will cause enormous social problems.

Workflow techniques are being applied to formalize and share the multi-user applications that characterize cloud computing in a series of processes. Workflow is a technique that describes a complex task with interdependencies as multiple processes instead of a single task or a task that is repeatedly performed as a flow.

In order to efficiently execute the tasks described in the above workflow in the cloud service, a scheduling technique for properly allocating the tasks of the workflow to virtualized resources is essential. Because of this length, the difference in execution cost depends on the scheduling technique. In other words, in the cloud computing environment, the workflow scheduling technique becomes an important factor for system efficiency, and an efficient workflow scheduling strategy can save costs not only for users but also for service providers.

Workflow scheduling algorithms have been proposed in grid computing and cluster environments, and classified into static and dynamic algorithms. Static algorithms estimate parameters such as task execution time or computational cost. Dynamic algorithms make scheduling decisions only when each task is ready to run. Most dynamic algorithms perform better than static algorithms.

However, since the cloud is dynamically set up for a wide range of scalability and user requirements, and many users can simultaneously use workflow applications in the cloud, the proposed algorithms are not suitable for the cloud computing environment.

Many existing workflow scheduling algorithms suggest scheduling for a single workflow, and these scheduling algorithms do not take into account multiple workflows that are characteristic of cloud computing, and focus on a single QoS parameter such as runtime and cost.

For example, Rank-HF (Highest Rank First) and Rank-HYBD (Rank Hybrid) are known as algorithms previously proposed for workflow scheduling.

Rank-HF sets the rank value in the task pool and sends it to the scheduler, and the scheduler sorts the task by the rank value and sends it to the executor, and the rank value of the task is made so that the higher rank task in the task pool becomes the priority. .

Rank-HYBD checks whether a task in the taskpool is a multi-workflow; if it is a single workflow, it is the same as Rank-HF; if it is a multi-workflow, the rank is determined by the taskpool and sent to the scheduler; This low task is sorted and sent to the executor, and the ranking of the task is made so that tasks with lower ranking values in the task pool are given priority.

That is, Rank-HYBD is configured to rank all prepared tasks, and determine which tasks are scheduled first according to the rank. However, if any new workflows with lower computational costs are submitted in the middle of execution, the tasks with higher computational costs will not receive resources or services until the tasks of the lower computational costs are completed. As a result, there is a risk of unnecessary long processing time or more system availability.

And in cloud computing, service evaluation is different for users and service providers. For example, the user side of cloud computing is concerned about how much they are willing to pay and are allocated computing resources that match their QoS requirements, while the provider side of cloud computing is concerned with the maximum benefit from providing computing resources. There is this.

Therefore, in the cloud computing environment, it is necessary to consider the total processing time of the user side (Makespan) and the availability of the entire system of the service provider side, and additionally, the service recovery average time (MTTR) as a parameter for reliable measurement of the service. You need to consider Mean Time To Recovery.

SUMMARY OF THE INVENTION The present invention has been made in view of the above, and is a workflow scheduling technique for multi-workflow applications with various QoS parameters and prioritization by cost value with throughput using expected execution time to suit the cloud computing environment. It is an object of the present invention to provide a cost-based scheduling method and system for multiple workflows in the cloud computing that can provide.

A cost-based scheduling system for multiple workflows in cloud computing according to an embodiment of the present invention includes a task pool, a scheduler, and an executor. Submitting a workflow with requirements involves assigning the appropriate services for workflow processing and scheduling service tasks according to the QoS requirements and cloud environment.

The task pool manages multiple workflows for dynamic cloud services, and when a workflow having a QoS requirement is submitted from a user, the task pool calculates characteristics such as estimated execution time and cost of all tasks and queues the tasks. Save to cue sheet.

The scheduler schedules tasks in a cue sheet according to a multiple workflow scheduling strategy.

The executor selects an optimal resource or service to execute the tasks in the cue sheet, and feeds back the task completion status to the task pool when the service of the randomly selected task is successfully completed.

The task pool removes the corresponding task when the task completion status is fed back from the executor.

In the cloud computing according to an embodiment of the present invention, a cost-based scheduling method for multiple workflows calculates characteristics such as estimated execution time and cost of all tasks when any new workflow is submitted from the user to the task pool in the standby state. Save the tasks to a cuesheet in the taskpool, the first task in the cuesheet is submitted to the scheduler, and the scheduler repeatedly recalculates and sorts the priorities with the calculated cost of all tasks, Checking if the service is empty, assigning the task to the appropriate service, and when the service completes, the executor involves feeding back the completion status to the task pool and removing the task from the cuesheet.

The scheduler schedules the tasks with the lowest cost, and preferentially schedules the task with the higher latency value if the cost values are the same.

The task pool checks whether any subtasks (child tasks) are preparing each time the executor has a task completion feedback, and then submits them to the scheduler.

The task pool submits information on characteristics of the task, such as expected execution time, cost, and delay time, to the scheduler together with the task.

Whenever there is a service available in the scheduler and the information of the task is calculated and waiting in the task pool, the task is repeatedly performed to sort all tasks by recalculating priorities.

According to the cost-based scheduling method and system for multi-workflow in cloud computing according to an embodiment of the present invention, the cost is scheduled in a low priority, if the cost value is the same, the delay time in the order (submitted order) Since tasks with high values are scheduled in priority order, a significant improvement in mean total process time (Mean Makespan), mean availability (Mean Availability), and mean time to recovery (MTTR) is obtained.

Therefore, the total processing time, which is the response time on the user side, is reduced and the availability is improved on the service provider side, thereby maximizing the profit in terms of the cost of the system.

In addition, according to the cost-based scheduling method and system for multiple workflows in the cloud computing according to an embodiment of the present invention, both the user and the supplier can achieve the effect of improving the execution time and reducing the waiting time.

1 is a block diagram schematically illustrating a workflow scheduling structure to which a cost-based scheduling system for multiple workflows is applied in cloud computing according to an embodiment of the present invention.
2 is a block diagram schematically illustrating a cost-based scheduling system for multiple workflows in cloud computing according to an embodiment of the present invention.
Figure 3 is a graph showing the average total processing time as a result of testing multiple workflows using an embodiment of the present invention, Rank-HF algorithm, Rank-HYBD algorithm.
Figure 4 is a graph showing the average availability as a result of testing multiple workflows using an embodiment of the present invention, Rank-HF algorithm, Rank-HYBD algorithm.
5 is a graph showing a service recovery average time as a result of testing multiple workflows using an embodiment of the present invention, the Rank-HF algorithm, and the Rank-HYBD algorithm.

Next, a preferred embodiment of a cost-based scheduling method and a system for multiple workflows in cloud computing according to the present invention will be described in detail with reference to the accompanying drawings.

First, a cost-based scheduling system for multiple workflows in cloud computing according to an embodiment of the present invention includes a task pool 22, a scheduler 24, and an executor 26.

1 illustrates a workflow scheduling structure, in which a cost-based scheduling system (CMWS) 20 for multiple workflows in cloud computing according to an embodiment of the present invention is provided between a workflow 10 and a cloud resource 30. Indicates that it is located.

The cost-based scheduling system 20 (CMWS) according to an embodiment of the present invention allocates appropriate services for workflow processing and QoS when users submit a workflow having a Quality of Service (QoS) requirement. Service tasks are scheduled according to requirements and cloud environment.

The task pool 22 manages multiple workflows 10 for dynamic cloud services, and when a workflow 10 having QoS requirements is submitted from a user, expected execution time and costs of all tasks, etc., are submitted. Compute the properties of and save the tasks to a cue sheet.

The scheduler 24 schedules tasks in a cue sheet according to a multiple workflow scheduling strategy.

The executor 26 selects an optimal resource or service to execute the tasks in the cue sheet, and feeds back the task completion status to the task pool 22 when the service of the randomly selected task is successfully completed.

The task pool 22 removes the corresponding task when the task completion status is fed back from the executor 26.

Once the workflows 10 are submitted to the taskpool 22, the launcher 26 may apply a scheduling algorithm to the tasks in the taskpool 22 even though the tasks belong to another workflow 10. In addition, the task pool 22 is capable of scheduling multiple workflows 10 that are dynamically submitted to the cloud.

When the workflows 10 are submitted to the task pool 22 when ready to run, they are executed in a limited manner in accordance with the executor 26.

Next, another embodiment of the present invention for performing cost-based scheduling for multiple workflows by applying a cost-based scheduling system (CMWS) for multiple workflows in the cloud computing according to an embodiment of the present invention configured as described above The cost-based scheduling method for multiple workflows in the cloud computing will be described.

First, when any new workflow 10 is submitted from the user to the task pool 22 in the wait state (for example, set to "0"), the characteristics such as estimated execution time and cost of all tasks are calculated. And stores the tasks in the cue sheet of the task pool 22.

The cost Cost (Task _i ) value for the tasks of each workflow 10 is calculated using Equation 1 below.

는 각 태스크의 평균 예상 실행시간을 나타낸다.In Equation 1, Tput (Task _i ) represents the respective throughput for the task of each workflow 10, Length (Task _i ) represents the size of each task,

Represents the average estimated execution time of each task.

In an embodiment of the present invention, all prepared tasks are scheduled and allocated to resources or services according to characteristics such as expected execution time and cost defined in Equation (1).

The first task in the cue sheet of the task pool 22 is then submitted to the scheduler 24.

The scheduler 24 repeatedly calculates and sorts priorities based on the calculated costs of all tasks (the cost value calculated through Equation 1).

The launcher 26 checks which service is empty, i.e., the service to which the task is not assigned and assigns the task to the appropriate service, i. Feedback to the task pool 22.

When the completion status is fed back from the executor 26, the task pool 22 removes the task from the cue sheet.

The task pool 22 checks whether any subtasks (child tasks) are preparing each time there is completion feedback of each task from the executor 26, and then submits them to the scheduler 24. Also perform.

In addition, the task pool 22 submits information on characteristics such as expected execution time, cost, and delay time of the task to the scheduler 24 together with the task.

The scheduler 24 schedules the tasks with the lowest cost, and preferentially schedules the task with the higher latency value if the cost values are the same.

In this case, the delay time represents the time from the time entered into the task pool 22 to the present, and when the delay time value is large, it means that the service is delayed longer.

Whenever there is a service available in the scheduler 24 and the information of the task is calculated and waiting in the task pool 22, the task of repeatedly recalculating the priority and sorting all tasks is repeatedly performed.

And the running task is set to "1", for example.

If there is no service that can complete the task in the executor 26, the task is inserted into another service, and all services are configured to execute only one task at the same time.

The completed task is returned to the task pool 22 by setting it to "2", for example, and the task pool 22 is configured to remove the task.

Next, performance evaluation and cost-based scheduling system (CMWS) for multiple workflows and existing algorithms Rank-HF and Rank-HYBD using a simulator considering cloud computing environment in cloud computing according to an embodiment of the present invention, Describe the process of performing the analysis.

First, the simulation environment is configured as shown in Table 1 below.

division Specification server  Intel SR1630HGP Processor  Intel QuadCore Xeon W3430 (2.40GHz) Processor cash  8MB L3 Cashe System bus  1333 MHz  Chipset  Intel Chipset 3420 Memory  2GB DDR3 800/1066 / 1333MHz Memorry 6DIMN Hard disk  500 GB SATA2 7200 rpm RAID  Intel SATA RAID supports RAID levels 0, 1 Drive bay  3 * 3.5 "Hotswap SATA Drive Bays OS  CentOS-5.4 * 86_64 (kernel 2.6.18-164) GCC  gcc version 4.1.2 20080704

And for the simulation, the cloud environment is set up 20 services, all services are configured to run a single task at the same time, 10 to 100 users are used to actually 10, 15, 20, 30, 50, 80 and 100 workflow applications were generated at the same time, 100 times for each algorithm. That is, performance evaluation was performed by applying each algorithm with the maximum workflow of about 100 ~ 1,200 tasks.

In addition, the size of the workflow 10 (the number of tasks in the workflow) uses a workflow group consisting of three to nine tasks, and is configured to generate four kinds of workflows randomly according to the workflow application scenario. .

The size of each task occurs randomly. The size of each task in the workflow is generated up to 50 to 1,000 units, and the maximum size is set to 1,000 units.

In addition, it is assumed that the time required to perform a service having a task size of 100 units is set to 1 second and 1,000 units to 10 seconds, and that the arrival interval follows the Poisson distribution, and that each task has an arrival interval of 200 units. do.

In addition, we used mean mean make time, mean availability, and mean time to recovery (MTTR).

The mean total processing time (Mean Makespan) refers to the total processing time or execution time when the workflow application enters the system, is scheduled, receives a service, and is executed. This time includes waiting time, which is a time waiting for service.

Mean availability refers to the average availability that a workflow can access resources or services in the cloud system. It is expressed as following Equation 2.

In Equation 2, UpTime refers to the total time that a service is available, Period of Time means a period of time (period for executing a task), MeanServiceTime represents an average time required for a service, and TaskRTime is used to execute a task. I = 1, ..., n denotes the number of tasks in the workflow.

The mean time to recovery (MTTR) means the average time that the system has not operated due to a failure and is a basic parameter of system reliability.

The service recovery average time (MTTR) is expressed by Equation 3 below.

In Equation 3, summation of TimeServiceRepair means total time for restoring a service, TotalNumber of ServiceRepair means total number for restoring a service, and WaitingTime means waiting time, i = 1, ..., n represents the number of tasks in the workflow.

Table 2 and FIG. 3 show the results of evaluating the performance of the mean total processing time (Mean Makespan) of the embodiment of the present invention (CMWS), Rank-HF, and Rank-HYBD.

division W10 W15 W20 W30 W50 W80 W100
Rank-HF Scenario 1 15.14 28.63 30.20 47.85 81.31 126.45 147.51 Scenario 2 21.71 30.89 36.10 54.22 86.88 147.84 172.64 Scenario 3 22.00 29.13 36.77 55.06 94.72 155.98 184.54 Scenario 4 23.40 29.89 39.18 56.74 93.12 154.78 182.88 Average 20.56 29.64 35.56 53.47 89.01 146.26 171.89
Rank-
HYBD Scenario 1 18.00 28.17 29.93 45.93 79.86 124.06 147.54 Scenario 2 20.29 25.78 32.13 51.61 85.41 141.52 162.35 Scenario 3 22.50 30.00 37.77 51.88 92.69 151.58 175.76 Scenario 4 20.60 29.89 37.73 54.74 87.65 149.40 175.18 Average 20.35 27.71 34.39 51.04 86.40 141.64 165.21
CMWS
(The present invention)
Examples) Scenario 1 14.43 23.58 24.53 39.11 68.60 108.79 112.47 Scenario 2 20.86 27.67 31.71 46.78 74.20 124.56 143.84 Scenario 3 20.00 27.53 34.23 50.18 82.44 136.11 156.63 Scenario 4 21.70 25.44 36.18 48.61 80.26 134.15 158.73 Average 19.25 26.06 31.66 46.17 76.38 125.90 142.92

Referring to Tables 2 and 3, when a workflow application is scheduled using a Rank-HF algorithm, a Rank-HYBD algorithm, and one embodiment (CMWS) according to the present invention, the total processing time and scenarios for each scenario The average total processing time of Rank-HF and Rank-HYBD algorithm is smaller as the number of workflows increases, but it can be seen that the Rank-HYBD algorithm has a smaller processing time on average than the Rank-HF algorithm. According to the exemplary embodiment (CMWS), as the number of workflows increases, the difference occurs more than the two algorithms.

That is, in the case of the embodiment according to the present invention (CMWS), as the number of workflows increases, it is possible to confirm that the processing time is significantly reduced on average compared to the conventional algorithms Rank-HF and Rank-HYBD. For example, in the case of one embodiment (CMWS) according to the present invention, an average of about 14% is improved compared to the Rank-HF algorithm and an average of about 10% is improved compared to Rank-HYBD.

Table 3 and FIG. 4 below show the evaluation results for average availability.

division W10 W15 W20 W30 W50 W80 W100
Rank-HF Scenario 1 70.06 55.16 39.13 31.75 21.74 14.72 11.50 Scenario 2 52.08 35.22 32.29 23.40 14.20 8.55 6.87 Scenario 3 48.98 37.40 31.38 19.42 11.98 7.80 6.37 Scenario 4 51.04 34.06 26.40 20.04 13.27 8.04 6.49 Average 55.54 40.46 32.30 23.65 15.30 9.78 7.81
Rank-
HYBD Scenario 1 61.32 57.63 42.91 35.63 23.84 15.19 11.18 Scenario 2 53.53 44.51 32.40 25.12 13.94 10.00 7.64 Scenario 3 45.31 33.81 30.07 20.91 12.89 7.86 6.93 Scenario 4 48.11 36.08 33.46 21.94 13.98 7.99 6.68 Average 52.07 43.01 34.71 25.90 16.16 10.26 8.11
CMWS
(The present invention)
Examples) Scenario 1 78.30 61.69 53.90 41.24 28.22 19.30 16.28 Scenario 2 63.60 46.19 44.63 30.60 20.43 12.11 10.26 Scenario 3 62.89 47.29 39.94 27.06 16.94 11.01 8.92 Scenario 4 61.48 48.70 33.53 27.19 17.44 10.84 8.87 Average 66.57 50.97 43.00 31.52 20.76 13.32 11.08

Referring to Tables 3 and 4, it can be seen that the availability decreases in all three cases as the number of workflows increases. It can be seen that as the number of tasks increases, the contention rate of services or resources increases, and that the service is not ready to run immediately.

Availability represents the resource availability of the system, so availability for each workflow or scenario is meaningless, so we only calculate and evaluate the average availability of the entire workflow.

Referring to Table 3 and FIG. 4, in the case of one embodiment (CMWS) according to the present invention, availability is generally higher than that of other algorithms, and an average of about 32% is higher than that of Rank-HF algorithm, compared to Rank-HYBD. The service availability is high on average 27%.

Table 4 and FIG. 5 show evaluation results of the MTTR.

division W10 W15 W20 W30 W50 W80 W100
Rank-HF Scenario 1 2.52 6.42 7.73 13.44 23.99 38.82 45.80 Scenario 2 2.12 3.87 4.71 7.72 12.99 22.75 26.71 Scenario 3 1.67 2.46 3.24 5.27 9.44 15.93 18.95 Scenario 4 1.62 2.46 3.48 5.47 9.23 15.79 18.74 Average 1.98 3.80 4.79 7.98 13.91 23.32 27.55
Rank-
HYBD Scenario 1 3.57 5.75 7.98 12.98 23.74 38.44 45.94 Scenario 2 2.26 3.26 4.42 7.29 12.80 21.95 25.26 Scenario 3 1.67 2.51 3.35 4.99 9.26 15.64 18.19 Scenario 4 1.48 2.53 3.38 5.14 8.71 15.39 18.18 Average 2.25 3.51 4.78 7.60 13.63 22.86 26.89
CMWS
(The present invention)
Examples) Scenario 1 2.00 5.03 5.49 10.28 19.70 32.67 34.22 Scenario 2 1.86 3.02 3.78 6.20 10.60 18.68 21.27 Scenario 3 1.89 2.01 2.68 4.37 7.81 13.58 15.80 Scenario 4 1.31 1.80 2.95 4.31 7.68 13.38 16.00 Average 1.77 2.97 3.73 6.29 11.45 19.58 21.82

The MTTR is a parameter for measuring the reliability of a service and represents an average time of service failure maintenance time, and represents an average time for tasks of each workflow to wait for service.

Referring to Table 4 and FIG. 5, in the case of one embodiment (CMWS) according to the present invention, the service recovery average time (MTTR) value is monotonically increased in 80 and 100 workflows. This means that the time for tasks to wait for service is less than the Ramk-HF and Rank-HYBD algorithms.

In particular, in the case of the embodiment according to the present invention (CMWS) it can be seen that the average service failure time is about 19% shorter than the Rank-HF algorithm, about 18% compared to Rank-HYBD.

As can be seen from Tables 2 to 4 and FIGS. 3 to 5, in the mean total process time (Mean Makespan), mean availability, and service recovery mean time (MTTR), one embodiment according to the present invention In the case of the CMWS, the average total processing time is lower than that of the Rank-HF and Rank-HYBD algorithms, and the system availability is high.

And the waiting time for tasks in the workflow to receive the cloud service is about 20% shorter than the Rank-HF and Rank-HYBD algorithm in one embodiment according to the present invention (CMWS).

In the case of the CMWS according to the present invention, the total processing time, which is the response time, is reduced in terms of the cloud service user, and the remaining time in the system is reduced by increasing the availability of the cloud service system in the cloud service provider. You can benefit from the cost.

In addition, the increase of the MTTR value in the system increases the memory usage and the CPU usage, thereby increasing the overall system load, but in the case of the CMWS according to the present invention, the user and the provider Both aspects can reduce the load on the system by improving run time and reducing latency.

In the above description of a preferred embodiment of a cost-based scheduling method and a system for multiple workflows in cloud computing according to the present invention, the present invention is not limited thereto, but is within the scope of the claims and the specification and the accompanying drawings. Various modifications can be made and this is also within the scope of the present invention.

10-workflow, 20-cost-based scheduling system, 22-task pool
24-Scheduler, 26-Launcher, 30-Cloud Resources

Claims (12)

It includes a task pool, a scheduler, and an executor.
When users submit a workflow with a Quality of Service (QoS) requirement, they are assigned to allocate services for workflow processing and to schedule service tasks according to QoS requirements and cloud environment.
The task pool manages multiple workflows for dynamic cloud services, and calculates expected execution time, cost, and latency characteristics of all tasks when a workflow with QoS requirements is submitted from the user. ),
The scheduler schedules tasks in the cuesheet according to a multiple workflow scheduling strategy,
The executor selects an optimal resource or service to execute the tasks in the cue sheet, feeds back the task completion status to the task pool when the service of the randomly selected task is completed successfully,
The task pool removes the corresponding task when the task completion status is fed back from the executor,
Cost (Task _i ) value for the tasks of each workflow is

Where Tput (Task _i ) represents the respective throughput for the task in each workflow, and Length (Task _i ) represents the size of each task,

Cost-based scheduling system for multiple workflows in cloud computing. The method according to claim 1,
The task pool is a cost-based scheduling system for multiple workflows in the cloud computing to submit information on the expected execution time, cost, latency characteristics of the task with the scheduler. delete The method according to claim 1,
Wherein said scheduler schedules the tasks with the lowest cost, and preferentially schedules the task with the higher latency value if the cost values are the same. The method according to claim 1,
The task pool checks whether any subtasks (child tasks) are preparing whenever the executor has a task's completion feedback, and then submits the multiple workflows in cloud computing to perform the process of submitting to the scheduler. Cost-based scheduling system. The method according to claim 1,
Whenever there is a service available in the scheduler and the information of the task is calculated and waiting in the task pool, multiple workflows are performed in the cloud computing which repeatedly performs the scheduling task of re-computing and sorting all tasks. Cost-based scheduling system. When a new workflow is submitted from the user to the task pool in the waiting state, it calculates the expected execution time, cost, and latency characteristics of all tasks, stores the tasks in the queue pool of the task pool,
The first task in the cuesheet is submitted to the scheduler, and the scheduler repeatedly performs the task of recalculating and sorting the priorities with the calculated costs of all tasks,
The launcher checks the service to which no task is assigned and assigns the task to the service,
When the service completes, the launcher includes feeding back the completion status to the taskpool and removing the task from the cuesheet,
Cost (Task _i ) value for the tasks of each workflow is

Where Tput (Task _i ) represents the respective throughput for the task in each workflow, Length (Task _i ) represents the size of each task,

Cost-based scheduling for multiple workflows in cloud computing. The method of claim 7,
The task pool is a cost-based scheduling method for multiple workflows in the cloud computing to submit information on the expected execution time, cost, latency characteristics of the task with the scheduler. delete The method of claim 7,
The scheduler schedules the tasks with the lowest cost, and preferentially schedules the task with the higher latency value if the cost values are the same. The method of claim 7,
The task pool checks whether any subtasks (child tasks) are preparing whenever the executor has a task's completion feedback, and then submits the multiple workflows in cloud computing to perform the process of submitting to the scheduler. Cost-based Scheduling Method. The method of claim 7,
Whenever there is a service available in the scheduler and the information of the task is calculated and waiting in the task pool, multiple workflows are performed in the cloud computing which repeatedly performs the scheduling task of re-computing and sorting all tasks. Cost-based Scheduling Method.

Images