KR101670460B1 - Apparatus and method for controlling scheduling of workflow based on cloud - Google Patents

Apparatus and method for controlling scheduling of workflow based on cloud Download PDF

Info

Publication number
KR101670460B1
KR101670460B1 KR1020150053600A KR20150053600A KR101670460B1 KR 101670460 B1 KR101670460 B1 KR 101670460B1 KR 1020150053600 A KR1020150053600 A KR 1020150053600A KR 20150053600 A KR20150053600 A KR 20150053600A KR 101670460 B1 KR101670460 B1 KR 101670460B1
Authority
KR
South Korea
Prior art keywords
workflow
time
scheduling
job
task
Prior art date
Application number
KR1020150053600A
Other languages
Korean (ko)
Other versions
KR20160124278A (en
Inventor
윤찬현
하윤기
Original Assignee
한국과학기술원
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 한국과학기술원 filed Critical 한국과학기술원
Priority to KR1020150053600A priority Critical patent/KR101670460B1/en
Publication of KR20160124278A publication Critical patent/KR20160124278A/en
Application granted granted Critical
Publication of KR101670460B1 publication Critical patent/KR101670460B1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
    • H04L41/5003—Managing SLA; Interaction between SLA and QoS
    • H04L41/5006—Creating or negotiating SLA contracts, guarantees or penalties
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
    • H04L41/5003—Managing SLA; Interaction between SLA and QoS
    • H04L41/5019—Ensuring fulfilment of SLA
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/50—Network service management, e.g. ensuring proper service fulfilment according to agreements
    • H04L41/5029—Service quality level-based billing, e.g. dependent on measured service level customer is charged more or less

Abstract

A cloud-based workflow scheduling method according to the present invention comprises: a first step of receiving a workflow service request from a user including a processing time (D) of a workflow and quality constraints on which cost constraints are set; A second step of calculating a minimum processing expected time (CT) for executing a workflow of a job; and a step of, when a time condition larger than the calculated minimum processing expected time of the processing time is satisfied, A third step of calculating a load ratio of each of the jobs, and a third step of setting a penalty cost caused by a violation of a service level agreement (SLA), and defining a cost of the user, a set penalty cost, A fourth step of establishing a cost model in which the cost model is established; It may include a fifth step of executing the scheduling.

Description

[0001] APPARATUS AND METHOD FOR CONTROLLING SCHEDULING OF WORKFLOW BASED ON CLOUD [0002]

The present invention relates to a technique for scheduling a workflow, and more particularly, to a method and apparatus for scheduling a workflow in a cloud computing environment, And more particularly, to a cloud-based workflow scheduling apparatus and method suitable for satisfying quality of service constraints (QoS constraints) of the same user.

In general, a cloud service broker (CSB) has a service level agreement (SLA) on the scope of QoS for users and workflow processing to provide cloud service consumers (users) with workflow processing services to provide. [Blake, M. Brian, and David J. Cummings. "Workflow composition of service level agreements." Services Computing, 2007. SCC 2007. IEEE International Conference on. IEEE, 2007.], [Wu, Linlin, Saurabh Kumar Garg, and Rajkumar Buyya. "SLA-based admission control for a software-as-a-service provider in cloud computing environments." Journal of Computer and System Sciences 78.5 (2012): 1280-1299.]

Therefore, the CSB side should adopt a workflow scheduling technique to efficiently utilize the QoS constraints of users. Here, the scheduling problem of the workflow is the uncertainty about the processing time of each task, And there is a difference in scheduling problem due to uncertainty due to allocation to a limited resource set. [Greg ㅄ orio Baggio Tramontina, Jacques Wainer, Clarence Ellis, "Applying scheduling techniques to minimize the number of late jobs in workflow systems." Proceeding of 2004 ACM Symposium on Applied Computing, Nicosia, Cyprus, 3, 2004, pp. 1396? 1403.]

In particular, in a cloud environment, CSBs should be careful to prevent SLA violations, because even if they try to run the same task on the same resource type, performance changes are severe. [Kim, Byungsang (2013). "A Study on Cost Adaptive Cloud Resource Broker System for Bio-Workflow Computing." Ph. D. Dissertation, Korea Advanced Institute of Science of Technology]

Meanwhile, the phased workflow scheduling scheme is a method of avoiding the workflow execution path prediction by mixing the static workflow scheduling technique and the dynamic workflow scheduling technique and coping with the uncertainty of the workflow scheduling.

However, the existing step-by-step workflow scheduling technique is effective in coping with changes in cloud resource performance. However, if the performance degradation of the cloud resource that handles a specific task of the workflow is severe, the additional cost to avoid the SLA violation There is a problem that requires time.

U.S. Published Patent Application No. 2011-0231899 (Published on September 22, 2011) United States Patent No. 8725546 (Notification date: 2014. 05. 13.)

[1] Topcuoglu, Haluk, Salim Hariri, and Min-you Wu. "Performance-effective and low-complexity task scheduling for heterogeneous computing." Parallel and Distributed Systems, IEEE Transactions on 13.3 (2002): 260-274. [2] Sakellariou, Rizos, et al. "Scheduling workflows with budget constraints." Integrated Research in GRID Computing. Springer US, 2007. 189-202. [3] Xiao, Zhijiao, and Zhong Ming. "A method of workflow scheduling based on colored Petri nets." Data & Knowledge Engineering 70.2 (2011): 230-247. [4] Kim, Daesun (2014). "Adaptive Workflow Scheduling Scheme Based on the Colored Petri-Net Model in Cloud." Master's thesis, Korea Advanced Institute of Science of Technology. [5] Van der Aalst, Wil MP. "The application of Petri nets to workflow management." Journal of circuits, systems, and computers 8.01 (1998): 21-66. [6] Kelley Jr, James E. "Critical-path planning and scheduling: Mathematical basis." Operations Research 9.3 (1961): 296-320.

The present invention relates to a method and system for effectively utilizing a completion time of a workflow process, which is a QoS constraint set by a user, in processing a workflow on a cloud service broker (CSB) side In this paper, we propose a cloud-based workflow scheduling scheme that divides each work that constitutes a workflow and distributes them to multiple cloud resources for scheduling.

In addition, the present invention enables a workflow process to be executed without violating a service level agreement (SLA) even when the completion time of the process is insufficient, thereby reducing the cost required for workflow processing through job division scheduling. We propose a workflow scheduling technique.

The problems to be solved by the present invention are not limited to those mentioned above, and another problem to be solved by the present invention can be clearly understood by those skilled in the art from the following description will be.

According to one aspect of the present invention, there is provided a method for processing a workflow, comprising: a first step of receiving, from a user, a workflow service request including a processing time (D) of a workflow and quality constraints for which a cost constraint is set; A second process of calculating a minimum processing expected time (CT) for execution; and a second step of, when a time condition larger than the calculated minimum processing expected time is satisfied, A third step of calculating a load ratio, a method of setting a penalty cost caused by a violation of a service level agreement (SLA), a cost model defined by a user's equity cost, the set penalty cost, And a fifth step of executing scheduling when the profit based on the cost model satisfies a profit condition that is larger than a predetermined value, Based workflow scheduling methodology.

The method of the present invention is characterized in that, when the time condition is not satisfied or the profit condition is not satisfied, the job division policy is applied to each job until the threshold division degree, and then the second to fifth processes are executed Process.

The second step of the present invention includes the steps of calculating a minimum processing time of the workflow for each task by transmitting a token from the end place of the workflow in a reverse direction, And calculating the minimum processing expected time.

The minimum processing time of the present invention can be determined by the critical path of the workflow.

The critical path of the present invention is a path having a longest processing completion time determined based on an average processing completion time for an individual task of the workflow, and can be expressed as a set of places and a set of transitions.

The minimum processing expected time of the present invention can be calculated as a sum of minimum processing times of jobs belonging to the set of transitions.

The method of the present invention may further include returning the workflow service request to the user when the time condition is not satisfied or the profit condition is not satisfied even after applying the job division policy.

The method of the present invention may further include the step of terminating the virtual machine when the execution of the scheduling is completed and returning the virtual machine to the corresponding cloud infrastructure provider.

According to another aspect of the present invention, there is provided a workflow interface for receiving a workflow service request from a user including a processing time (D) of a workflow and quality constraints for which a cost constraint is set, Calculating a minimum processing expected time (CT) for execution and calculating a load ratio of each task when a time condition (D > CT) in which the processing time is greater than the minimum processing expected time is satisfied, A workflow analyzer for instructing the application of the job division policy when the job division is not satisfied, and a penalty point calculation unit for setting a penalty cost caused by the breach of the service level agreement (SLA) and calculating a penalty cost of the user, (Pf> n) in which the profit (Pf) by the established cost model is larger than a predetermined value (n) is established A workflow partitioning policy manager for instructing the execution of scheduling when the profit and loss conditions are satisfied and for requesting application of a work partitioning policy when the profit condition is not satisfied; A task scheduler that executes scheduling based on the contents of task profiling when a condition that a result of mapping each task to D> CT is satisfied and a condition of Pf> n is satisfied in the cost model, And requesting each of the cloud infrastructure providers to create and rent a performance virtual machine required for execution of the task, and when the execution of the scheduling task is completed by the task scheduler, the virtual machine is automatically terminated and returned to the corresponding cloud infrastructure provider Based workflow scheduling apparatus that includes a machine manager.

The workflow analyzer of the present invention calculates the minimum processing time of the workflow for each job by transmitting the tokens from the end place of the workflow in the reverse direction to calculate the minimum processing time of the workflow based on the sum of the calculated minimum processing times To calculate the minimum processing expected time.

The job scheduler of the present invention performs job division repeatedly if the profit Pf 'updated by the job division is not greater than the predetermined value, and if the number of jobs divided by the threshold division degree is generated, If the value is not larger than the value, the job scheduling can be performed by taking a form having the smallest loss among the divided types of jobs.

The task scheduler of the present invention is characterized in that, even though the task division policy is applied up to the threshold division degree, when the time condition of D> CT is not satisfied or the profit condition of Pf> n is not satisfied, A flow service request can be returned to request a workflow service request based on a new quality constraint.

The virtual machine manager of the present invention may store the work profile results of the work completion time of the virtual machine in the work profile repository.

The present invention estimates the processing completion time when a processing (execution) request of a workflow is received from a user, and when the processing time of the work flow requested by the user is received, a virtual machine (VM) (QoS) guaranteeing region for the existing scheduling method by using a penalty point function when the predicted processing completion time is insufficient, the service level agreement (SLA) is applied to the change of the performance of the cloud resource Can be satisfied satisfactorily.

FIG. 1 is a conceptual diagram of a workflow execution service providing system suitable for applying a cloud-based workflow scheduling apparatus according to the present invention.
FIG. 2 is a block diagram of a cloud-based workflow scheduling apparatus according to an embodiment of the present invention. Referring to FIG.
3 is a flowchart illustrating a main process of providing a cloud-based workflow scheduling service according to an embodiment of the present invention.
Figure 4 is an illustration of the structure of a workflow that can be refined by the user.
5 is a diagram illustrating an example of an inverse token processing process.
FIG. 6 is an exemplary diagram illustrating a forward token processing process.
7 is a diagram showing an example of an average execution profiling result of a task t i with a critical partition degree of 8;

First, the advantages and features of the present invention, and how to accomplish them, will be clarified with reference to the embodiments to be described in detail with reference to the accompanying drawings. While the present invention has been described with reference to exemplary embodiments, it is to be understood that the invention is not limited to the disclosed exemplary embodiments, but, on the contrary, It will be understood by those of ordinary skill in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined by the appended claims.

In the following description of the present invention, detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. It is to be understood that the following terms are defined in consideration of the functions of the present invention, and may be changed according to intentions or customs of a user, an operator, and the like. Therefore, the definition should be based on the technical idea described throughout this specification.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

FIG. 1 is a conceptual diagram of a workflow execution service providing system suitable for applying a cloud-based workflow scheduling apparatus according to the present invention.

1, a workflow execution service providing system may include a user terminal 110, a cloud service broker 120 and a plurality of cloud infrastructure providers 130, 140 and 150, and the cloud infrastructure providers 130, 140 and 150 may include a plurality of virtual machines (VM) 131 and 132 or 141 and 142 or 151 and 152, respectively.

The user terminal 110 may be a wired or wireless communication terminal used by a user (consumer) to receive (consume) a cloud service. The user terminal 110 may include a workflow structure and a workflow process A message for a workflow service request including quality constraints such as time constraints and cost constraints, and transmitting the message to the cloud service broker 120 via a network not shown.

Here, the structure of the workflow indicates the phases including the processing order of the work constituting the work flow, the type of the work, and the threshold partitioning degree (t i {n} , i: the type of operation and n: the critical partitioning degree). As an example, as shown in FIG. 4, the user can refine the structure of the workflow, where T1 corresponds to the task that is executed first, T2 and T4 can be executed after task T1 is completed, and T3 is T2 And T5 can be executed after the completion of the operations of T3 and T4.

Next, the cloud service broker 120 replaces the user (consumer) on behalf of the user (consumer) for the added value of the cloud service between the cloud service consumer (user) and a plurality of cloud infrastructure providers (cloud infrastructure providers) (Or an intermediary broker, a joining blocker) that provides workflow processing services.

In other words, the cloud service broker 120 may refer to a cloud-based workflow scheduling apparatus according to the present invention. The cloud service broker 120 may be implemented by a plurality of cloud infrastructure providers 130, 140, 150 in consideration of cost (cost) It is possible to provide a function of renting and utilizing the virtual machines 131, 132, 141, 142, 151, and 152 optimally and providing a workflow execution service to a user (consumer).

For example, the cloud service broker 120 may be configured to allow a user 110 to send a constraint and cost constraint on the structure of the workflow and the workflow processing time via the workflow interface 210 (workflow service request) (VM) to execute an individual task, and at the same time, the workflow processing is completed within the processing time set by the user so as not to violate the service level agreement (SLA) .

To this end, the cloud service broker 120, which may be defined as a cloud-based workflow scheduling apparatus according to the present invention, may include a configuration as shown in FIG. 2 as an example.

Although FIG. 1 includes three cloud infrastructure providers, this is only an exemplary illustration for the purpose of illustration, and the present invention is not necessarily limited thereto. Of course, tens or even tens to hundreds of cloud infrastructure providers may be configured as infrastructures.

FIG. 2 is a block diagram of a cloud-based workflow scheduling apparatus according to an embodiment of the present invention. The workflow interface 210 includes a workflow analyzer 220, a workflow division policy manager 230, A repository 240, a job scheduler 250 and a virtual machine manager 260, and so on.

Referring to FIG. 2, a cloud service broker can manage information on an average operation time according to a virtual machine type (VM type) through a job profiling repository 240 for scheduling, And VT j = (C j , H j , M j ) as a combination of parameters relating to computing resources with unchanging characteristics. Here, C j represents the core number of the CPU, H j represents the clock speed of the CPU, and M j represents the memory size, respectively.

The leasing cost for the virtual machine type j may be represented by P j , and the job profiling storage 240 may store the job profiling result for the execution completion time in various virtual machine types j. Here, the rental cost is information provided by each cloud infrastructure provider through the virtual machine manager 260 described later.

First, the workflow interface 210 receives a workflow service request from the user terminal 110 via a network (not shown), that is, a constraint on the processing time of the workflow set according to the user interface and a quality constraint And receiving a workflow service request including a workflow structure and delivering the received request to the workflow analyzer 220.

Next, the workflow analyzer 220 calculates (estimates) a minimum processing estimated time (CT) for executing a workflow of each job according to a service request when a workflow service request is transmitted from the workflow interface 210 Calculates the load ratio of each task when the time condition set by the user is greater than the calculated minimum processing time (CT), and when the time condition is not met, the workflow division policy manager 230) to instruct the application of the job division policy.

That is, the workflow analyzer 220 performs the reverse token processing on the workflow structure requested to be executed by the user. For example, the workflow may be defined as W = (P, T, A). Here, the place P = {p 1 , p 2 , ..., p k } represents the possible state of the workflow phase, and the transition T = {t 1 , t 2 , ... ., t k} is used to describe the average completion time and cost required for the processing of each job, an arc (arc) a = {<p i, t i + 1>, <p i + 1, t i + 2 >} It becomes a set connecting place and transition.

Herein, the token (token) is used to schedule the workflow by applying the technique of the present invention to the phase of the workflow depicted in Patri net, and when the application of one step to any task t i is terminated, i to p i + 1 , and controls the scheduling.

The reverse token processing is performed, for example, by the procedure shown in FIG. 5, which will be described in detail as follows.

That is, the processing time of the workflow is calculated by transferring the token from the end place of the workflow phase depicted in Patri net in the reverse direction, and then compared with the processing time D of the workflow requested by the user. At this time, the minimum processing time of the workflow is determined by the critical path of the workflow, which has the longest processing completion time determined based on the average processing completion time for the individual work of the workflow And can be expressed as a set of places (P c ) and a set of transitions (T c ) belonging to the critical path as shown in Equation (1) below.

[Equation 1]

CP = { Pc , Tc }

At this time, the minimum processing expected time (CT) of the workflow can be obtained as the sum of the minimum processing times of jobs belonging to Tc as shown in the following Equation (2).

&Quot; (2) &quot;

Figure 112015036972838-pat00001

At the same time, the workflow analyzer 220 calculates the load ratio r (t i ) of each job as shown in the following equation (3).

&Quot; (3) &quot;

Figure 112015036972838-pat00002

Meanwhile, a cloud service broker (CSB) according to the present invention means a workflow management system using a virtual machine provided by a public cloud infrastructure service, and a user has a consideration of a workflow execution request (or a workflow service request) To the cloud service broker, which is a budget for the cloud service broker to provide the workflow service to the user.

The cloud service broker then uses the budget to pay for renting virtual machines from a public cloud infrastructure provider (a cloud infrastructure service provider).

Next, the workflow partitioning policy manager 230 determines whether the time condition in which the processing time set by the user is greater than the minimum processing expected time CT is unsatisfied and the application of the job partitioning policy is instructed from the workflow analyzer 220 , We use the degree of violation of the service level agreement (SLA) set by the user in the penalty function technique to set the penalty cost caused by the SLA violation and use them to calculate the virtual The cost model is established (established) by the following equation (4) as the difference between the leasing cost and the penalty cost of the machine. Here, the penalty cost may be stored in the job profiling store 240.

&Quot; (4) &quot;

Pf = B - C l - C p

In Equation (4), Pf represents the profit, B represents the cost paid by the user to use the service, C l represents the penalty cost, and C p represents the cost required to operate the virtual machines .

That is, the workflow partitioning policy manager 230 uses the established cost model to determine how the cloud service broker will respond to the user's cloud service request as follows.

For example, if the result value shown in Equation (3) is larger than n, that is, if profit occurs, the workflow processing is immediately performed. If the result is less than n, the resultant value of workflow scheduling And proceeds to process according to the result of the workflow scheduling when profit is generated.

In other words, the workflow partitioning policy manager 230 determines the processing time (D) <the minimum processing time (CT) as a result of mapping each job to the virtual machine showing the best performance when the job division policy is not applied. Or in the cost model of the cloud service broker shown in Equation (4), Pf < n applies the task division policy up to the critical degree for each task. Here, n is a predetermined value and may indicate "0 ".

Here, a divisible task among the tasks constituting the workflow divides the data requiring processing in the task into an arbitrary form, distributes the data to the plurality of computing nodes, It can be defined as a task that can shorten the processing time when a series of programs to be executed in the original task are executed in the same manner.

In order to reduce the complexity of workflow scheduling considering work partitioning, a critical partitioning degree is defined. When a partitionable job is equally divided into sub-tasks of 2 squares so as to have the same completion time, The number of sub-tasks that can be included in the job.

For example, in the case where the threshold division degree of task A is 8, task A can be divided into three times in total, and the number of lower tasks is 8 at that time. When an arbitrary task t i is in a state capable of being divided into at most n, it is denoted by t i {n} , and the critical division degree cd (t i ) can be expressed as shown in Equation (5) below.

&Quot; (5) &quot;

cd (t i ) = n

At this time, there are various methods of job division, but in the present embodiment, a method of accurately dividing each job into half in order to reduce the complexity of scheduling and reduce labor required for job profiling has been exemplified. That is, the task t i has a critical division degree of n times the power of 2.

Then, the penalty point cost C 1 can be subdivided as shown in Equation (6) below.

&Quot; (6) &quot;

Figure 112015036972838-pat00003

In Equation (6), Cl is the sum of the rental costs C vmi of various virtual machine types (VM types) necessary for the workflow processing as the rental costs of the virtual machines. This can be expressed as the product of the virtual machine usage time ut vmi required for the task processing and the usage time init vmi required for the virtual machine operation multiplied by the unit cost per unit time of the virtual machine type.

In addition, the penalty cost (C p ) and the degree of SLA violation (SV) that are caused by violation of the service level agreement (SLA) that occurs when the workflow can not be processed within the minimum processing time (CT) 7 can be defined as follows.

&Quot; (7) &quot;

Figure 112015036972838-pat00004

Figure 112015036972838-pat00005

Next, the task scheduler 250 maps the tasks to the virtual machine showing the best performance when the task division policy is applied up to the threshold division degree, the time condition of D> CT is satisfied, and the above- When the profit condition of Pf > n is satisfied in the cost model of the cloud service broker shown in Equation 4, the workflow service request (workflow processing request) of the user is accepted and based on the contents of the job profiling, And executes the scheduling while moving the token in the forward direction in the same manner as described above.

At this time, the job scheduler 250 repeatedly performs job division if the profit Pf 'updated by the job division is not larger than the predetermined value n, and if the number of jobs divided by the threshold division degree is generated, If the value is not larger than the predetermined value, the job scheduling is performed by taking the form having the smallest loss among the job division types.

At this time, the virtual machine type mapped to the individual task is the sum of the actual execution times of the tasks on the critical path

Figure 112015036972838-pat00006
, Then the sum of the actual execution times of the tasks on the partial path
Figure 112015036972838-pat00007
, And the result of profiling the work in virtual machine type k is
Figure 112015036972838-pat00008
, The following expression (8) is satisfied.

&Quot; (8) &quot;

Figure 112015036972838-pat00009

FIG. 7 is a graph showing average execution profiling results of a task t i having a threshold division degree of 8 when three types (VT 1 , VT 2 , VT 3 ) of virtual machine types utilized by a cloud service broker (CSB) Fig.

In addition, although the task scheduler 250 has applied the task division policy up to the threshold partitioning degree, when the time condition of D> CT is not satisfied or the profit condition of Pf> n is not satisfied, Requesting the user to re-request the workflow service (re-creation request) by newly creating a quality restriction on the processing time and cost constraints necessary for processing the work flow, Functionality.

Finally, the virtual machine manager 260 requests the heterogeneous cloud infrastructure providers 130, 140, and 150 to create and lease virtual machines of the performance required for task execution, and executes the scheduling tasks by the task scheduler 250 It is possible to provide a function of automatically terminating the virtual machine when it is completed and returning the virtual machine to the cloud infrastructure provider so as to minimize the cost caused by resource leasing.

In addition, the virtual machine manager 260 may provide a function of storing the monitoring result of the virtual machine that is performing the job, that is, the job profile result of the job completion time, in the job profile storage 240.

Next, a series of processes for providing a cloud-based workflow scheduling service to users using the workflow scheduling apparatus of the present invention having the above-described configuration will be described in detail.

3 is a flowchart illustrating a main process of providing a cloud-based workflow scheduling service according to an embodiment of the present invention.

3, the workflow interface 210 receives a workflow service request from the user terminal 110 via the network, that is, a restriction on the processing time D of the workflow set according to the user interface, And a workflow service structure including a quality constraint and a workflow structure are received and transmitted to the workflow analyzer 220 (step 302).

Next, the workflow analyzer 220 calculates (calculates) the minimum processing expected time (CT) for executing the workflow of each job according to the service request (step 304), and determines whether the processing time D set by the user is It is checked whether a time condition (D > CT) larger than the calculated minimum processing expected time (CT) is satisfied (step 306).

That is, the workflow analyzer 220 calculates the minimum processing time of the workflow for each job by transmitting the tokens in the reverse direction from the end place of the workflow phase, and calculates the minimum processing time Where the minimum processing time is determined by the critical path of the workflow and the critical path is determined based on the average processing completion time for the individual work of the workflow It can be expressed as a set of places and a set of transitions as paths with the longest processing completion time.

Then, the minimum processing expected time is calculated as the sum of the minimum processing times of jobs belonging to the set of transitions.

As a result of the check in step 306, when the condition of D> CT is satisfied, the workflow analyzer 220 calculates the load ratio of each job (step 308).

Thereafter, the workflow segmentation policy manager 230 uses the degree of violation of the service level agreement (SLA) set by the user as a penalty function technique to set a penalty cost caused by the SLA violation, A cost model is set (step 310) as shown in Equation (4) based on the difference between the sum of the rental cost of the virtual machine and the penalty point cost in the budget paid by the user in the processing.

Next, the workflow partitioning policy manager 230 checks whether the profit (Pf) by the established cost model is greater than a predetermined value (n) is satisfied (step 312).

When the condition of Pf > n is satisfied as a result of the check in step 312, the job scheduler 250 executes job scheduling, i.e., accepts a user's workflow service request (workflow processing request) (Step 314), while moving the token in the forward direction based on the content of the token.

Thereafter, in the virtual machine manager 260, when the execution of the scheduling task by the task scheduler 250 is completed (step 316), the virtual machine manager 260 automatically terminates the virtual machine and returns it to the corresponding cloud infrastructure provider (step 318) The scheduling service is terminated.

Here, the virtual machine manager 260 may store the monitoring result for the virtual machine that is performing the job, that is, the job profile result for the job completion time, in the job profile storage 240.

On the other hand, if the condition of D> CT is not satisfied or the condition of Pf> n is not satisfied as a result of the check in step 312, the process proceeds to step 320 In step 320, a task division policy is applied to each task up to the threshold division degree.

When the condition of the progress D> CT and the condition of Pf> n are satisfied, the processes of the steps 322 to 330 are performed in the same manner as the steps 304 to 312 described above. The process proceeds to the above-described step 314 to execute subsequent processes (scheduling execution, virtual machine termination and return, etc.).

On the other hand, when the condition of D> CT is not satisfied as a result of the check in step 324, or when the condition of Pf> n is not satisfied as a result of the check in step 330, that is, The task scheduler 250 rejects the user's workflow service request (returns the workflow service request), and if the time condition of D> CT is not satisfied or the profit condition of Pf> n is not satisfied , A quality constraint is newly created for the processing time and the cost constraint necessary for the processing of the workflow, and a request is made to the user to request the workflow service again (step 332).

It will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the present invention as defined by the following claims. It is easy to see that this is possible. That is, the embodiments disclosed in the present invention are not intended to limit the scope of the present invention but to limit the scope of the present invention.

Therefore, the scope of protection of the present invention should be construed in accordance with the following claims, and all technical ideas within the scope of equivalents should be interpreted as being included in the scope of the present invention.

Claims (13)

A workflow scheduling method performed by a virtual machine,
A first step of receiving, from a user, a workflow service request including a processing time (D) of a workflow and a quality constraint on which a cost constraint is set;
A second step of calculating a minimum processing expected time (CT) for executing a workflow of each job according to a service request,
A third step of calculating a load ratio of each of the jobs based on the minimum processing expected time when a time condition in which the processing time is larger than the calculated minimum processing expected time is satisfied;
A fourth step of setting a penalty cost caused by a violation of a service level agreement (SLA), establishing a cost model defined by a user's equity cost, the set penalty cost, and the operating cost of the virtual machine,
A fifth step of executing scheduling when a profit condition in which the profit by the cost model is larger than a predetermined value is satisfied,
Based workflow scheduling method.
The method according to claim 1,
The method comprises:
Executing the second to fifth processes after applying the job division policy up to the threshold division degree for each job when the time condition is not satisfied or the profit condition is not satisfied
Based workflow scheduling method.
The method according to claim 1,
In the second process,
Transmitting a token from the end place of the workflow in a reverse direction to calculate a minimum processing time of the workflow for each task;
Calculating a minimum processing expected time based on a sum of minimum processing times of the respective jobs;
Based workflow scheduling method.
The method of claim 3,
The minimum processing time may be,
Is determined by the critical path of the workflow
A cloud-based workflow scheduling method.
5. The method of claim 4,
The critical path includes:
A path having a longest processing completion time determined based on an average processing completion time for the individual work of the workflow, the path being represented by a set of places and a transition
A cloud-based workflow scheduling method.
6. The method of claim 5,
The minimum processing expected time is a time
Is calculated as the sum of the minimum processing times of jobs belonging to the set of transitions
A cloud-based workflow scheduling method.
The method according to claim 1,
The method comprises:
And returning the workflow service request to the user when the time condition is not satisfied or the profit condition is not satisfied even after applying the job division policy
Based workflow scheduling method.
The method according to claim 1,
The method comprises:
A process of terminating the virtual machine when the execution of the scheduling is completed and returning the virtual machine to the cloud infrastructure provider
Based workflow scheduling method.
A workflow interface for receiving a workflow service request from a user, the workflow service request including a processing time (D) of a workflow and a quality constraint for which a cost constraint is set;
Calculating a minimum processing expected time (CT) for executing a work flow of each job according to a service request, and when a time condition (D> CT) in which the processing time is longer than the minimum processing expected time is satisfied, A workflow analyzer that calculates a ratio and directs the application of a task partitioning policy when the time condition is not satisfied;
Establishing a cost model defined by the user's equity cost, the set penalty cost and the operating cost of the virtual machine, and establishing a cost model based on the established cost model A workflow partition for instructing the execution of scheduling when profit condition (Pf > n) where profit Pf is larger than a predetermined value n is satisfied, Policy managers,
When mapping the respective tasks to the virtual machine showing the best performance when the task division policy is applied up to the threshold division degree satisfies the condition of D> CT and the condition of Pf> n in the cost model is satisfied A task scheduler that executes scheduling based on the contents of the task profiling,
And requesting each of the cloud infrastructure providers to create and rent a performance virtual machine required for execution of the task, and when the execution of the scheduling task is completed by the task scheduler, the virtual machine is automatically terminated and returned to the corresponding cloud infrastructure provider Machine Manager
Based workflow scheduling device.
10. The method of claim 9,
The workflow analyzer includes:
Transferring the token from the end place of the workflow in the reverse direction to calculate a minimum processing time of the workflow for each task and calculating the minimum processing expected time based on the calculated sum of the minimum processing times of each task
Cloud-based workflow scheduling device.
10. The method of claim 9,
The job scheduler,
If the revenue Pf 'updated by the job division is not larger than the predetermined value, the job division is repeatedly performed. If the number of jobs divided by the threshold division degree is generated but the profit is not larger than the predetermined value, The task scheduling is performed by taking the relatively smallest loss form
Cloud-based workflow scheduling device.
10. The method of claim 9,
The job scheduler,
If the time condition of D> CT is not satisfied or the profit condition of Pf> n is not satisfied even though the task division policy is applied up to the threshold division degree, the user's workflow service request is returned to the new Request a workflow service request based on quality constraints
Cloud-based workflow scheduling device.
10. The method of claim 9,
Wherein the virtual machine manager comprises:
Storing the job profile results for the job completion time of the virtual machine in the job profile storage
Cloud-based workflow scheduling device.
KR1020150053600A 2015-04-16 2015-04-16 Apparatus and method for controlling scheduling of workflow based on cloud KR101670460B1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
KR1020150053600A KR101670460B1 (en) 2015-04-16 2015-04-16 Apparatus and method for controlling scheduling of workflow based on cloud

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
KR1020150053600A KR101670460B1 (en) 2015-04-16 2015-04-16 Apparatus and method for controlling scheduling of workflow based on cloud

Publications (2)

Publication Number Publication Date
KR20160124278A KR20160124278A (en) 2016-10-27
KR101670460B1 true KR101670460B1 (en) 2016-11-01

Family

ID=57247274

Family Applications (1)

Application Number Title Priority Date Filing Date
KR1020150053600A KR101670460B1 (en) 2015-04-16 2015-04-16 Apparatus and method for controlling scheduling of workflow based on cloud

Country Status (1)

Country Link
KR (1) KR101670460B1 (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR102139054B1 (en) * 2019-05-22 2020-07-29 주식회사 크라우드웍스 Method for measuring unit price of project based on crowd sourcing
KR102139141B1 (en) * 2019-05-29 2020-07-29 주식회사 크라우드웍스 Method for selecting worker accoring to feature of project based on crowd sourcing

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8931038B2 (en) 2009-06-19 2015-01-06 Servicemesh, Inc. System and method for a cloud computing abstraction layer
KR101350755B1 (en) * 2011-01-14 2014-01-10 대전대학교 산학협력단 Cost Based Scheduling Algorithm for Multiple Workflow in Cloud Computing and System of The Same
KR20140102478A (en) * 2013-02-14 2014-08-22 한국전자통신연구원 Workflow job scheduling apparatus and method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
'QoS 보장 컴퓨팅 클라우드 프레임워크', 연구보고서, 대전대학교(2011)
'클라우드 컴퓨팅에서 결정테이블을 이용한 워크플로우 스케줄링', 한국산업정보학회논문지 제17권제5호(2012)

Also Published As

Publication number Publication date
KR20160124278A (en) 2016-10-27

Similar Documents

Publication Publication Date Title
Garg et al. SLA-based virtual machine management for heterogeneous workloads in a cloud datacenter
Devi et al. Load balancing in cloud computing environment using improved weighted round robin algorithm for nonpreemptive dependent tasks
US10069693B1 (en) Distributed resource allocation
Hoenisch et al. Cost-efficient scheduling of elastic processes in hybrid clouds
Lucas‐Simarro et al. Cost optimization of virtual infrastructures in dynamic multi‐cloud scenarios
US20090241117A1 (en) Method for integrating flow orchestration and scheduling for a batch of workflows
US11573835B2 (en) Estimating resource requests for workloads to offload to host systems in a computing environment
US10360074B2 (en) Allocating a global resource in a distributed grid environment
Truong Huu et al. Joint elastic cloud and virtual network framework for application performance-cost optimization
Zoghi et al. Designing adaptive applications deployed on cloud environments
Moens et al. Cost-effective feature placement of customizable multi-tenant applications in the cloud
Bousselmi et al. Cloud services orchestration: A comparative study of existing approaches
Wei et al. Proactive virtualized resource management for service workflows in the cloud
US20200169602A1 (en) Determining allocatable host system resources to remove from a cluster and return to a host service provider
Biswas et al. An auto-scaling framework for controlling enterprise resources on clouds
Kumar et al. Time and cost optimization algorithm for scheduling multiple workflows in hybrid clouds
Wang et al. Kepler+ cometcloud: Dynamic scientific workflow execution on federated cloud resources
Genez et al. Using time discretization to schedule scientific workflows in multiple cloud providers
Rost et al. It's about time: On optimal virtual network embeddings under temporal flexibilities
Ramezani et al. An online fuzzy decision support system for resource management in cloud environments
Kecskemeti et al. Facilitating self-adaptable Inter-Cloud management
Ghasemi et al. A cost-aware mechanism for optimized resource provisioning in cloud computing
Genez et al. Time-discretization for speeding-up scheduling of deadline-constrained workflows in clouds
Ben Halima et al. Optimal business process deployment cost in cloud resources
JP7182836B2 (en) Automatic Diagonal Scaling of Workloads in Distributed Computing Environments

Legal Events

Date Code Title Description
A201 Request for examination
E902 Notification of reason for refusal
E701 Decision to grant or registration of patent right
GRNT Written decision to grant
FPAY Annual fee payment

Payment date: 20191001

Year of fee payment: 4