CN107070965B - Multi-workflow resource supply method under virtualized container resource - Google Patents

Multi-workflow resource supply method under virtualized container resource Download PDF

Info

Publication number
CN107070965B
CN107070965B CN201611199049.XA CN201611199049A CN107070965B CN 107070965 B CN107070965 B CN 107070965B CN 201611199049 A CN201611199049 A CN 201611199049A CN 107070965 B CN107070965 B CN 107070965B
Authority
CN
China
Prior art keywords
container
resource
workflow
resources
virtualized
Prior art date
Legal status (The legal status 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 status listed.)
Active
Application number
CN201611199049.XA
Other languages
Chinese (zh)
Other versions
CN107070965A (en
Inventor
李棕
钟积海
崔得龙
彭志平
柯文德
李启锐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong University of Petrochemical Technology
Original Assignee
Guangdong University of Petrochemical Technology
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 Guangdong University of Petrochemical Technology filed Critical Guangdong University of Petrochemical Technology
Priority to CN201611199049.XA priority Critical patent/CN107070965B/en
Publication of CN107070965A publication Critical patent/CN107070965A/en
Application granted granted Critical
Publication of CN107070965B publication Critical patent/CN107070965B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • 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
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • 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
    • G06F9/46—Multiprogramming arrangements
    • G06F9/48—Program initiating; Program switching, e.g. by interrupt
    • G06F9/4806—Task transfer initiation or dispatching
    • G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
    • G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/1023—Server selection for load balancing based on a hash applied to IP addresses or costs

Landscapes

  • Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention discloses a multi-workflow resource supply method under virtualized container resources, which schedules workflows by utilizing reinforcement learning, supplies resources, defines a resource utility index U, establishes a supply-demand relation between tasks running in a container cluster and the virtualized container resources at each resource scheduling time, and meets the requirement of multi-workflow resource generation with the container cluster as granularity by designing a reward function: the method not only ensures that the number and the types of container units in the container cluster meet the operation flow of the cloud workflow, but also avoids the workflow with different QoS requirements from violating the service level agreement, and improves the resource utilization rate of the whole container cluster. The state information of the tasks in each container cluster can be acquired in real time, and the workflow task allocation and the virtualization resource supply are coordinated with each other.

Description

Multi-workflow resource supply method under virtualized container resource
Technical Field
The invention relates to the field of cloud computing, in particular to a multi-workflow resource supply method under virtualized container resources.
Background
It is very difficult to perform workflow tasks and virtualized resource cooperative adaptive scheduling in a transient and variable cloud computing environment. For example, data centers of Amazon, IBM, microsoft and Yahoo all have hundreds of thousands of servers, the number of servers owned by Google is even more than 100 thousands, the number of various physical resources is larger after virtualization, physical nodes and virtualization units are down, dynamically added and removed, and the like, and the management technology is difficult and complex. As another example, taking a multi-layer Web services workflow as an example, the load change law due to an emergency is often unpredictable.
Scheduling of various types of cloud workflow tasks across multiple processing units has proven to be an NP-complete challenge from a task optimization allocation perspective. From the perspective of resource optimization supply, on one hand, energy consumption needs to be considered in virtual unit placement, namely, the number of activated physical machines and used network devices is reduced, and at the moment, virtual unit placement can be abstracted to a boxing problem, which is an NP complete difficult problem; on the other hand, the transmission of data between virtual units needs to be considered, that is, the use of network bandwidth is reduced, and at this time, the virtual unit placement can be abstracted to a secondary allocation problem, which is also a complete problem of NP. The existing cloud workflow scheduling focuses on workflow task allocation under fixed virtualization resources, or focuses on elastic resource supply under workflow load change, or focuses on how to integrate the existing workflow management system into a cloud platform, and the workflow task allocation and the virtualization resource supply cannot be coordinated with each other.
Disclosure of Invention
The invention aims to solve the defects of the prior art and provides a workflow resource supply method which can mutually cooperate workflow task allocation and virtualization resource supply, and the technical scheme is as follows:
a multi-workflow resource supply method under a virtualized container resource adopts a container resource generation strategy based on reinforcement learning, and comprises the following steps:
defining a state space: representing a state space by using a quintuple S ═ WR, RA, AW, IM and PJ, wherein WR is the workload of the workflow tasks to be scheduled, RA is the resource available time, AW is the total workload of the workflow tasks in the waiting queue, IM is the number of idle container resources, and PJ is the proportion of submitting the workflow tasks by each user in the queue;
defining an action space: the action space comprises two actions of a workflow task to be distributed and the number of requested resources;
setting a reward function Re＝λeW+(1-λe) U, wherein λe∈[0,1]Is a control coefficient; w is the task response rate:
Figure BDA0001188645620000021
the execution time is the execution time of the workflow task, the waiting time is the waiting time of the workflow task, and U is the resource utility index:
Figure BDA0001188645620000022
[Tk,...,Tk+1]indicates the resource supply decision time, PkIs represented by [ Tk,...,Tk+1]Available container resources in a time container cluster, fnRepresents TNThe sum of the execution time of the workflow task at the moment;
setting a reward function upper limit value RuLower limit value RlRetention range Rm～Rn；
Selecting a workflow task to be executed from the action space, executing the selected task, detecting and acquiring a reward function Rε；
If the reward function RεGreater than RuThen, in the subsequent execution process of the task, the virtualized container resource in the cloud platform is added, and if the reward function R is usedεLess than RuThen, in the subsequent execution process of the task, the virtualized container resources of the cloud platform are reduced, and if the reward function R is usedεAt Rm～RnWithin range, virtualized container resources in the cloud platform are kept unchanged.
The invention utilizes reinforcement learning to schedule the workflow and supply resources, defines a resource utility index U, establishes a supply-demand relation between tasks running in a container cluster and virtualized container resources at each resource scheduling time, and meets the requirement of generating the multi-workflow resources with the container cluster as granularity by designing a reward function: the method not only ensures that the number and the types of container units in the container cluster meet the operation flow of the cloud workflow, but also avoids the workflow with different QoS requirements from violating the service level agreement, and improves the resource utilization rate of the whole container cluster. The state information of the tasks in each container cluster can be acquired in real time, and the workflow task allocation and the virtualization resource supply are coordinated with each other.
Preferably, the present invention further includes deploying the virtualized container resource, specifically including:
clustering based on the minimal cut intra-cluster virtualization container hierarchy;
optimizing network flow by using a local search algorithm;
optimizing the placement of the virtualized containers using an optimal matching algorithm: when a newly created virtualization container is placed, searching is carried out in sequence from the first used physical machine, the physical machine which is matched with the virtualization container is found for placing, and a new physical machine is enabled only when all the used physical machines cannot accommodate the virtualization container.
The best matching means that the residual available resources (including CPU, memory and bandwidth) of the currently enabled physical machine meet the requirements of the newly created virtualized container on the resources, and the residual resources most possibly meet the requirements of the newly created virtualized container on the resources next time or are the least.
The method for optimizing the network flow by using the local search algorithm specifically comprises the following steps: selecting a virtualized container which generates a congested link and has the maximum flow on the basis of the minimum hierarchical clustering result by taking the maximum link utilization rate or the hot spot link number as an objective function, randomly exchanging the virtualized container with containers under left and right neighbor switches, and then calculating the objective function: if the objective function value is decreased, accepting the exchange; if not, refusing exchange, repeating in turn until circulating to the set iteration times.
Compared with the prior art, the invention has the beneficial effects that:
the invention utilizes reinforcement learning to schedule the workflow and supply resources, defines a resource utility index U, establishes a supply-demand relation between tasks running in a container cluster and virtualized container resources at each resource scheduling time, and meets the requirement of generating the multi-workflow resources with the container cluster as granularity by designing a reward function: the method has the advantages that the number and the types of container units in the container cluster are ensured to meet the operation flow of cloud workflow, the condition that the workflow with different QoS requirements violates a service level agreement is avoided, the resource utilization rate of the whole container cluster is improved, the state information of tasks in each container cluster can be acquired in real time, and the task allocation and the virtualized resource supply of the workflow are mutually coordinated.
Drawings
FIG. 1 is a system model schematic of the present invention;
FIG. 2 is a graph illustrating container hierarchical clustering based on minimal cut according to the present invention.
Detailed Description
The invention is described in further detail below with reference to the figures and examples.
Example (b):
a system model for multi-workflow resource supply under virtualized container resources is shown in fig. 1, and resource-level supply corresponds requirements of users submitting workflow tasks to specific resources, so that the performance of applications is optimal, and the utilization rate of resources is improved as much as possible. The workflow under the cloud environment is composed of a series of subtasks, and at the moment, the supply of the resource level is to select proper available resources to complete the creation of the virtual unit, namely the generation problem of the virtual unit; secondly, how to reduce the number of active physical machines and the occupation of network bandwidth, namely the problem of placing virtual units.
The embodiment adopts the following technical scheme to achieve the effect of mutual cooperative matching of resource supply and workflow tasks:
a multi-workflow resource supply method under a virtualized container resource adopts a container resource generation strategy based on reinforcement learning, and comprises the following steps:
defining a state space: representing a state space by using a quintuple S ═ WR, RA, AW, IM and PJ, wherein WR is the amount of the workflow tasks to be scheduled, RA is the resource available time, AW is the total amount of the workflow tasks in the waiting queue, IM is the number of idle container resources, and PJ is the proportion of submitting the workflow tasks by each user in the queue;
defining an action space: the action space comprises two actions of a workflow task to be distributed and the number of requested resources;
setting a reward function Re＝λeW+(1-λe) U, wherein λe∈[0,1]Is a control coefficient; w is the task response rate:
Figure BDA0001188645620000051
the execution time is the execution time of the workflow task, the waiting time is the waiting time of the workflow task, and U is the resource utility index:
Figure BDA0001188645620000052
[Tk,...,Tk+1]indicates the resource supply decision time, PkIs represented by [ Tk,...,Tk+1]Available container resources in a time container cluster, fnRepresents TNThe sum of the execution time of the workflow task at the moment;
setting a reward function upper limit value RuLower limit value RlRetention range Rm～Rn；
Selecting a workflow task to be executed from the action space, executing the selected task, detecting and acquiring a reward function Rε；
If the reward function RεGreater than RuThen, in the subsequent execution process of the task, the virtualized container resource in the cloud platform is added, and if the reward function R is usedεLess than RuThen, in the subsequent execution process of the task, the virtualized container resources of the cloud platform are reduced, and if the reward function R is usedεAt Rm～RnWithin range, virtualized container resources in the cloud platform are kept unchanged.
The embodiment further includes deploying the virtualized container resource, specifically including:
clustering based on the minimal cut intra-cluster virtualization container hierarchy;
optimizing network flow by using a local search algorithm;
optimizing the placement of the virtualized containers using an optimal matching algorithm: when a newly created virtualization container is placed, searching is carried out in sequence from the first used physical machine, the physical machine which is matched with the virtualization container is found for placing, and a new physical machine is enabled only when all the used physical machines cannot accommodate the virtualization container.
And G ═ (V, E) represents a DAG graph of the cloud workflow, wherein V represents a container cluster, E represents the traffic among containers in the cluster, and a node set is represented as
Figure BDA0001188645620000061
The set of edges is denoted as δ (Q). Then in graph G, one vertex of the edge is in set Q and the other vertex belongs to V \ Q when
Figure BDA0001188645620000062
Or Q ≠ V, the edges in δ (Q) constitute a cut set, denoted (Q, V \ Q). For each edge (i, j) E, there is a non-negative capacity Ci,j. And the capacity of a cut set can be defined as the sum of the capacities of each edge in the cut set, and can be expressed as: c (Q, V \ Q) ═ Σi,j∈δ(Q)C(i,j)。
Hierarchical clustering based on minimal cuts is to find a cut set with the smallest capacity in graph G. Taking fig. 2 as an example, the graph G may be represented by a binary tree T (V), where the left subtree TL is a node in Q, and the weight is the sum of the edge values in Q, W (TL) ═ Σi,j∈δ(Q)C (i, j); right subtree TR is a node of V \ Q, and the weight is sum W (TR) of edge values in V \ Q ═ Σi,j∈δ(Q)C (i, j), if W (TL)<W (TR), the left and right subtrees are swapped to ensure that traffic flow in the left subtree TL is always greater than the right subtree.
The method for optimizing the network flow by using the local search algorithm specifically comprises the following steps: selecting a virtualized container which generates a congested link and has the maximum flow on the basis of the minimum hierarchical clustering result by taking the maximum link utilization rate or the hot spot link number as an objective function, randomly exchanging the virtualized container with containers under left and right neighbor switches, and then calculating the objective function: if the objective function value is decreased, accepting the exchange; if not, refusing exchange, repeating in turn until circulating to the set iteration times.

Claims (3)

1. A multi-workflow resource supply method under a virtualized container resource is characterized in that a container resource generation strategy based on reinforcement learning is adopted, and the method comprises the following steps:
defining a state space: representing a state space by using a quintuple S ═ WR, RA, AW, IM and PJ, wherein WR is the workload of the workflow tasks to be scheduled, RA is the resource available time, AW is the total workload of the workflow tasks in the waiting queue, IM is the number of idle container resources, and PJ is the proportion of submitting the workflow tasks by each user in the queue;
defining an action space: the action space comprises two actions of a workflow task to be distributed and the number of requested resources;
setting a reward function Re＝λeW+(1-λe) U, wherein λe∈[0,1]For the control coefficient, W is the task response rate:
Figure FDA0001188645610000011
the execution time is the execution time of the workflow task, the waiting time is the waiting time of the workflow task, and U is the resource utility index:
Figure FDA0001188645610000012
[Tk,...,Tk+1]indicates the resource supply decision time, PkIs represented by [ Tk,...,Tk+1]Available container resources in a time container cluster, fnRepresents TNThe sum of the execution time of the workflow task at the moment;
setting a reward function upper limit value RuLower limit value RlRetention range Rm～Rn；
Selecting a workflow task to be executed from the action space, executing the selected task, detecting and acquiring a reward function Rε；
If the reward function RεGreater than RuThen, in the subsequent execution process of the task, the virtualized container resource in the cloud platform is added, and if the reward function R is usedεLess than RuThen, in the subsequent execution process of the task, the virtualized container resources of the cloud platform are reduced, and if the reward function R is usedεAt Rm～RnWithin range, virtualized container resources in the cloud platform are kept unchanged.
2. The method for supplying resources of multiple workflows under the virtualized container resource according to claim 1, further comprising deploying the virtualized container resource, specifically comprising:
clustering based on the minimal cut intra-cluster virtualization container hierarchy;
optimizing network flow by using a local search algorithm;
optimizing the placement of the virtualized containers using an optimal matching algorithm: when a newly created virtualization container is placed, searching is carried out in sequence from the first used physical machine, the physical machine which is matched with the virtualization container is found for placing, and a new physical machine is enabled only when all the used physical machines cannot accommodate the virtualization container.
3. The method for supplying resources of multiple workflows under the condition of virtualizing container resources according to claim 2, wherein the optimizing network traffic by using the local search algorithm specifically comprises: selecting a virtualized container which generates a congested link and has the maximum flow on the basis of the minimum hierarchical clustering result by taking the maximum link utilization rate or the hot spot link number as an objective function, randomly exchanging the virtualized container with containers under left and right neighbor switches, and then calculating the objective function: if the objective function value is decreased, accepting the exchange; if not, refusing exchange, repeating in turn until circulating to the set iteration times.
CN201611199049.XA 2016-12-22 2016-12-22 Multi-workflow resource supply method under virtualized container resource Active CN107070965B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611199049.XA CN107070965B (en) 2016-12-22 2016-12-22 Multi-workflow resource supply method under virtualized container resource

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611199049.XA CN107070965B (en) 2016-12-22 2016-12-22 Multi-workflow resource supply method under virtualized container resource

Publications (2)

Publication Number Publication Date
CN107070965A CN107070965A (en) 2017-08-18
CN107070965B true CN107070965B (en) 2021-02-19

Family

ID=59618941

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611199049.XA Active CN107070965B (en) 2016-12-22 2016-12-22 Multi-workflow resource supply method under virtualized container resource

Country Status (1)

Country Link
CN (1) CN107070965B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107908457B (en) * 2017-11-08 2020-03-17 河海大学 Containerized cloud resource allocation method based on stable matching
CN108829494B (en) * 2018-06-25 2020-09-29 杭州谐云科技有限公司 Container cloud platform intelligent resource optimization method based on load prediction
CN109639791A (en) * 2018-12-06 2019-04-16 广东石油化工学院 Cloud workflow schedule method and system under a kind of container environment
CN115022188B (en) * 2022-05-27 2024-01-09 国网经济技术研究院有限公司 Container placement method and system in electric power edge cloud computing network

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007080047A1 (en) * 2006-01-10 2007-07-19 International Business Machines Corporation Method and system for sending and processing electronic messages
CN103699446A (en) * 2013-12-31 2014-04-02 南京信息工程大学 Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method
CN105260230A (en) * 2015-10-30 2016-01-20 广东石油化工学院 Resource scheduling method for data center virtual machine based on segmented service level agreement
CN105897864A (en) * 2016-03-28 2016-08-24 东南大学 Scheduling method for cloud workflow

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2007080047A1 (en) * 2006-01-10 2007-07-19 International Business Machines Corporation Method and system for sending and processing electronic messages
CN103699446A (en) * 2013-12-31 2014-04-02 南京信息工程大学 Quantum-behaved particle swarm optimization (QPSO) algorithm based multi-objective dynamic workflow scheduling method
CN105260230A (en) * 2015-10-30 2016-01-20 广东石油化工学院 Resource scheduling method for data center virtual machine based on segmented service level agreement
CN105897864A (en) * 2016-03-28 2016-08-24 东南大学 Scheduling method for cloud workflow

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
IaaS环境下改进能源效率和网络性能的虚拟机放置方法;董健康;《通信学报》;20140131;第35卷(第1期);第1-10页 *

Also Published As

Publication number Publication date
CN107070965A (en) 2017-08-18

Similar Documents

Publication Publication Date Title
Abdulhamid et al. Fault tolerance aware scheduling technique for cloud computing environment using dynamic clustering algorithm
Awad et al. Enhanced particle swarm optimization for task scheduling in cloud computing environments
Wang et al. Adaptive scheduling for parallel tasks with QoS satisfaction for hybrid cloud environments
Mansouri et al. Cost-based job scheduling strategy in cloud computing environments
CN107070965B (en) Multi-workflow resource supply method under virtualized container resource
CN114138486A (en) Containerized micro-service arranging method, system and medium for cloud edge heterogeneous environment
Malik et al. Comparison of task scheduling algorithms in cloud environment
CN108270805B (en) Resource allocation method and device for data processing
Dutta et al. A smart job scheduling system for cloud computing service providers and users: modeling and simulation
Li et al. An effective scheduling strategy based on hypergraph partition in geographically distributed datacenters
Lagwal et al. Load balancing in cloud computing using genetic algorithm
CN106201701A (en) A kind of workflow schedule algorithm of band task duplication
Kao et al. Data-locality-aware mapreduce real-time scheduling framework
Delavar et al. A synthetic heuristic algorithm for independent task scheduling in cloud systems
Bansal et al. A multi-faceted optimization scheduling framework based on the particle swarm optimization algorithm in cloud computing
Pandey et al. A heuristic method towards deadline-aware energy-efficient mapreduce scheduling problem in Hadoop YARN
Shi et al. MapReduce short jobs optimization based on resource reuse
CN105404549A (en) Yarn architecture-based virtual machine scheduling system
CN114327811A (en) Task scheduling method, device and equipment and readable storage medium
Zhang et al. An energy-aware host resource management framework for two-tier virtualized cloud data centers
He et al. Energy-efficient framework for virtual machine consolidation in cloud data centers
Grami An energy-aware scheduling of dynamic workflows using big data similarity statistical analysis in cloud computing
Lee et al. A hierarchical scheduling strategy for the composition services architecture based on cloud computing
Mao et al. Optimal scheduling algorithm of MapReduce tasks based on QoS in the hybrid cloud
Biswas et al. A novel resource aware scheduling with multi-criteria for heterogeneous computing systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant