CN111209104A - Energy perception scheduling method for Spark application under heterogeneous cluster - Google Patents
Energy perception scheduling method for Spark application under heterogeneous cluster Download PDFInfo
- Publication number
- CN111209104A CN111209104A CN202010315850.6A CN202010315850A CN111209104A CN 111209104 A CN111209104 A CN 111209104A CN 202010315850 A CN202010315850 A CN 202010315850A CN 111209104 A CN111209104 A CN 111209104A
- Authority
- CN
- China
- Prior art keywords
- task
- spark
- tasks
- energy consumption
- machine
- 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.)
- Pending
Links
Images
Classifications
-
- 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
-
- 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
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- 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
- G06F9/54—Interprogram communication
- G06F9/546—Message passing systems or structures, e.g. queues
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2209/00—Indexing scheme relating to G06F9/00
- G06F2209/54—Indexing scheme relating to G06F9/54
- G06F2209/548—Queue
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management
Abstract
The invention discloses an energy perception scheduling method for Spark application under a heterogeneous cluster, which is from the perspective of a cloud service provider, meets the requirements of users, and simultaneously minimizes the energy consumption of a provider data center. The Spark application submitted by the user can be decomposed into a task sequence, the tasks are distributed to proper machines, and meanwhile, the task execution sequence is dynamically adjusted, so that the current task can meet the deadline while the adjustment cost is minimized. The invention optimizes the energy consumption of Spark application execution by realizing dynamic allocation of proper resources.
Description
Technical Field
The invention relates to an energy perception task scheduling method for Spark application in a heterogeneous cluster environment with hard deadline constraints, and belongs to the technical field of cloud computing resource scheduling.
Background
Over the past decade, cloud computing data centers have accounted for 2.4% of the total electricity used globally, and have produced an economic impact of $ 300 billion. The energy consumption of the data center increases by more than 15% every year, and the energy cost accounts for about 42% of the operation cost of the data center. Considering that the cost of hardware such as servers is still decreasing, it can be said that a large part of the cost of data centers in the near future is the cost of energy consumption. Therefore, it is essential that the data center minimize energy consumption when providing services to customers.
Task scheduling is critical to energy conservation in data centers. For example, In an AIS (All-In stratum) system, as long as one server still works, other idle servers In the cluster cannot be uniformly shut down, and a large amount of energy is wasted by the data center. If a reasonable scheduling strategy is used, the servers can be completed at almost the same time, so that on one hand, the task execution time can be shortened, and on the other hand, the servers can be closed uniformly as soon as possible, thereby reducing the energy consumption.
Spark default scheduling strategies are 2: the FIFO mode and the Fair share the mode scheduling strategy, but the two strategies lack overall control and allocation of resources, which causes much unnecessary overhead and waste of resources. For example, in the FIFO mode, the task in the Spark task stream is allocated to the cluster according to the initial task sequence, and then each node in the cluster may run with an overload on the high-energy-consumption node and the low-energy-consumption node is in an "idle" state during task execution, which causes a problem that the whole application cannot complete execution failure within the deadline or the energy consumption is too high.
Disclosure of Invention
The purpose of the invention is as follows: aiming at the scheduling optimization problem in the cloud computing environment, the energy-aware scheduling method of Spark application in heterogeneous clusters is provided by considering the benefit of maximizing the user level or the service provider level. The method is based on the Spark task scheduling problem based on energy perception under the data center heterogeneous cluster, the characteristics of Spark application workflow are considered, the Spark task scheduler is improved based on the existing Spark task scheduler to construct an energy consumption perception task, and Spark application is executed while the least energy is consumed under the condition of meeting SLA. Wherein SLA refers to the hard deadline of the application.
The technical scheme is as follows: an energy-aware scheduling method for Spark application under heterogeneous clusters comprises the following steps:
step 1, determining effective topological sorting of Spark workflow according to partial sorting property of Spark workflow, and generating a scheduling queue Q according to a sorting model;
step 2, taking out the tasks vi from the call queue Q in sequence, calculating the processing time of the tasks vi on each physical machine Mk to obtain the earliest completion time, and calculating the minimum scheduling length to obtain the exact relaxation time of the Spark application;
step 3, performing execution time estimation on all tasks, and performing sub-deadline division on the tasks according to the estimated task execution time and the deadline of the Spark application workflow;
step 4, allocating each task vi in the scheduling queue Q to heterogeneous resources, and calculating total energy consumption E;
step 5, after step 4 is executed, an initial scheduling solution is obtained, some slack time still exists in the machine execution process, particularly, a time block existing between two continuous tasks on the same machine is still idle, the frequency is reduced by using the DVFS, and the energy consumption is further reduced by reducing the slack time;
step 6, a Variable Neighborhood Descent (VND) -based task sequence adjusting method is adopted, a new task scheduling series is generated by exchanging task nodes in an initial task scheduling sequence Q0, and the step 2 is switched to obtain a resource allocation scheme with lower energy consumption;
step 7, if the task queue Q is empty, the method is ended; otherwise, go to step 4.
In step 1, for the partial order relationship between Job and Stage in Spark application, two aspects are considered: the Job layer considers and designs a corresponding sequencing rule to obtain a sequencing sequence of Job; the sequence of the Stage level refers to a sequencing sequence obtained by designing a corresponding sequencing rule in the same Job on one hand, and refers to a sequencing sequence of stages in different Jobs on the other hand.
The Spark workflow and the sequencing model in the step 1 are specifically represented as follows:
spark task stream: s { v1, v 2.,. vN } is a Task flow with partial order relation on all three levels of Job, Stage and Task;
the specific ranking models are three types as follows:
(1) calculating an Upward Rank value for each task, and arranging the tasks in an increasing order according to Rank (vi) obtained by calculation;
(2) calculating a descending Rank value for each task, and arranging the tasks in an increasing order according to Rank (vi) obtained by calculation;
(3) and calculating a Hybrid Rank value for each task, and arranging the tasks in a descending order according to the calculated Rank (vi).
The step 4 specifically comprises the following steps:
step 41, finding out machines meeting the task vi, and then adding acceptable machines of the task into a resource sorting queue PM;
step 42, if the resource sorting queue PM is not empty, it indicates that there is a machine that meets the condition available for allocation; sequencing the machines in the PM, sequentially dequeuing the sequenced physical machine queues to be matched with the tasks vi, and if the sub deadline of the tasks is not exceeded, distributing the tasks to the machines for execution;
if there are no machines that meet the sub-deadline constraint, then task replicas vi ' need to be created for tasks vi ' and the machine with the smallest earliest completion time is selected for each task service replica vi ', step 43.
The machine sequencing rule in step 42 specifically includes:
(1) energy consumption priority rules: in a machine resource queue PM, sequencing the resource queues in an increasing order according to energy consumption, and preferentially selecting a machine with lower energy consumption to be distributed to a task for execution;
(2) energy consumption performance priority rule: in a machine resource queue PM, sequencing the resource queues in an increasing order according to the energy consumption performance ratio; preferentially selecting machines with lower energy consumption performance to be allocated to tasks to be executed; the method does not take into account the effect of frequency on energy consumption, that is, all machines should be executed at maximum frequency;
(3) local priority principle: in the machine resource queue PM, firstly, whether the local data required by the task vi is on the same machine as the current task is judged, and a machine storing the local data is preferentially selected to be distributed to the task for execution.
In step 4, the heterogeneous resources and the total energy consumption are specifically:
heterogeneous resources refer to the heterogeneity of the frequency and number of cores of the CPUs of a cluster of machines in data.
The total energy consumption refers to the total energy consumption consumed by the machine cluster to successfully execute the Spark application within the deadline, and even if some machines are in an idle state, the consumed energy is also counted into the total energy consumption.
Has the advantages that: compared with Spark default technology, the energy perception scheduling method for Spark applications in heterogeneous clusters optimizes energy consumption of Spark application execution by realizing dynamic allocation of appropriate resources.
Drawings
FIG. 1 is a block diagram of a Spark task stream according to an embodiment of the present invention;
FIG. 2 is a block diagram of an embodiment of the present invention;
FIG. 3 is a flow chart upper portion of a method of implementing an embodiment of the present invention;
fig. 4 is a lower part of a flow chart of an implementation method of an embodiment of the invention.
Detailed Description
The present invention is further illustrated by the following examples, which are intended to be purely exemplary and are not intended to limit the scope of the invention, as various equivalent modifications of the invention will occur to those skilled in the art upon reading the present disclosure and fall within the scope of the appended claims.
As shown in fig. 1, the data center is assumed to include 3 types of machines in the embodiment: high frequency machines, medium frequency machines, low frequency machines. Spark workflow S = { v1, v2, v3, … vN }, where vi is the ith task of Spark workflow, deadline is D.
According to the energy perception scheduling method of Spark application in heterogeneous cluster environment, firstly, for the partial order relation of Job and Stage in Spark application, two aspects are considered: the Job layer considers and designs a corresponding sequencing rule to obtain a sequencing sequence of Job; the sequence of the Stage layer refers to a sequencing sequence obtained by designing a corresponding sequencing rule in the same Job on one hand, and refers to a sequencing sequence of stages in different Jobs on the other hand; then, task allocation is carried out, tasks are allocated to core of a proper machine according to the sequenced Job and Stage scheduling sequences, and the task allocation comprises three stages; the first stage is as follows: establishing an initial task scheduling sequence; the invention combines the existing algorithm to design a proper sequencing method to generate an initial scheduling sequence of tasks; and a second stage: after the initial task scheduling solution is established, some slack time still exists in the machine execution process, the DVFS is used for reducing the frequency, and the energy consumption can be further reduced by reducing the slack time; and a third stage: and adjusting the task sequence. In view of the fact that the optimal machine allocation scheme obtained through initial task scheduling sequence searching is probably a local optimal solution, in order to search out a global optimal solution as much as possible, a variable domain descent (VND) -based task sequence adjusting method is adopted, and a resource allocation scheme with minimized energy consumption is obtained by exchanging new task scheduling sequences generated by task nodes in a task sequence.
As shown in fig. 2 to 4, the specific implementation steps of the energy-aware scheduling method for Spark application in the heterogeneous cluster environment according to the embodiment of the present invention are as follows:
step s201, dividing Spark workflow into Job according to action operator, adding all Job into J' queue according to partial order relation;
step s202, updating Job in the queue J 'to the ready Job queue J, and deleting the Job added to the queue J from the queue J';
step S203, synchronizing step S201, dividing stages in Job according to wide dependence, and adding all stages into an S' queue according to partial order relation;
step S204, updating the Stage in the S 'queue to a ready Stage queue S, and deleting the Stage added to S from S';
step S205, updating all the tasks in the ready queue S to the task set T to be scheduled;
step s206, sorting the scheduling priorities defined by the tasks in the task set T in an increasing mode to obtain a task scheduling queue Q;
step s207, sorting the tasks in the task set T in an increasing manner according to the defined scheduling priority to obtain a task scheduling queue Q;
and step s208, judging whether the task scheduling queue Q is empty, if so, executing the step s214, and ending the method. If not, go to step s 209;
step S209 is to schedule the task according to steps S201-S208, so as to obtain the minimum scheduling length sl (g) of the Spark application.
Step s210, dividing the relaxation time of the Spark application program DS (G) = D-SL (G) proportionally according to the maximum depth h of G to obtain the relaxation time of the task vi; g represents the whole spark application;
step s211, allocating the tasks vi in the task sequence Q to heterogeneous machines, and calculating the energy consumption of the tasks executed on the machines;
step s212, calculating a time parameter of the task, and determining whether the task can be relaxed when being executed on the machine. If not, performing step s208, if slackable, performing step s 213;
step s213, using DVFS to reduce frequency, further reducing energy consumption by reducing relaxation time;
step s214, calculating the total energy consumption E of the whole Spark application.
Claims (6)
1. An energy-aware scheduling method for Spark application in heterogeneous cluster environment is characterized by comprising the following steps:
step 1, determining effective topological sorting of Spark workflow according to partial sorting property of Spark workflow, and generating a scheduling queue Q according to a sorting model;
step 2, taking out the tasks vi from the call queue Q in sequence, calculating the processing time of the tasks vi on each physical machine Mk to obtain the earliest completion time, and calculating the minimum scheduling length to obtain the exact relaxation time of the Spark application;
step 3, performing execution time estimation on all tasks, and performing sub-deadline division on the tasks according to the estimated task execution time and the deadline of the Spark application workflow;
step 4, allocating each task vi in the scheduling queue Q to heterogeneous resources, and calculating total energy consumption E;
step 5, after step 4 is executed, obtaining an initial scheduling solution, and reducing the frequency by using the DVFS;
step 6, a task sequence adjusting method based on variable neighborhood descending is adopted, a new task scheduling series is generated by exchanging task nodes in an initial task scheduling sequence Q0, and the step 2 is switched to obtain a resource allocation scheme with lower energy consumption;
step 7, if the task queue Q is empty, the method is ended; otherwise, go to step 4.
2. The energy-aware scheduling method for Spark applications in heterogeneous cluster environment according to claim 1, wherein in step 1, for the partial order relationship between Job and Stage in Spark applications, two aspects are considered: the Job layer considers and designs a corresponding sequencing rule to obtain a sequencing sequence of Job; the sequence of the Stage level refers to a sequencing sequence obtained by designing a corresponding sequencing rule in the same Job on one hand, and refers to a sequencing sequence of stages in different Jobs on the other hand.
3. The energy-aware scheduling method for Spark applications in a heterogeneous cluster environment according to claim 1, wherein the Spark workflow and the sequencing model in step 1 are expressed as follows:
spark task stream: s { v1, v 2.,. vN } is a Task flow with partial order relation on all three levels of Job, Stage and Task;
the specific ranking models are three types as follows:
firstly, calculating an Upward Rank value for each task, and arranging the tasks in an increasing order according to Rank (vi) obtained by calculation;
secondly, calculating a downstream Rank value for each task, and arranging the tasks in an increasing order according to Rank (vi) obtained by calculation;
thirdly, calculating a Hybrid Rank value for each task, and arranging the tasks in a descending order according to Rank (vi) obtained by calculation.
4. The energy-aware scheduling method for Spark applications in a heterogeneous cluster environment according to claim 1, wherein the step 4 specifically includes:
step 41, finding out machines meeting the task vi, and then adding acceptable machines of the task into a resource sorting queue PM;
step 42, if the resource sorting queue PM is not empty, it indicates that there is a machine that meets the condition available for allocation; sequencing the machines in the PM, sequentially dequeuing the sequenced physical machine queues to be matched with the tasks vi, and if the sub deadline of the tasks is not exceeded, distributing the tasks to the machines for execution;
if there are no machines that meet the sub-deadline constraint, then task replicas vi ' need to be created for tasks vi ' and the machine with the smallest earliest completion time is selected for each task service replica vi ', step 43.
5. The energy-aware scheduling method for Spark applications in a heterogeneous cluster environment according to claim 4, wherein the machine ordering rule in step 42 specifically includes:
energy consumption priority rules: in a machine resource queue PM, sequencing the resource queues in an increasing order according to energy consumption, and preferentially selecting a machine with lower energy consumption to be distributed to a task for execution;
energy consumption performance priority rule: in a machine resource queue PM, sequencing the resource queues in an increasing order according to the energy consumption performance ratio; preferentially selecting machines with lower energy consumption performance to be allocated to tasks to be executed;
local priority principle: in the machine resource queue PM, firstly, whether the local data required by the task vi is on the same machine as the current task is judged, and a machine storing the local data is preferentially selected to be distributed to the task for execution.
6. The energy-aware scheduling method for Spark applications in heterogeneous cluster environment according to claim 1, wherein in step 4, the heterogeneous resources and the total energy consumption are specifically: the heterogeneous resources refer to the heterogeneity of the frequency and the core number of the CPU of the machine cluster in the data; the total energy consumption refers to the total energy consumption consumed by the machine cluster to successfully execute the Spark application within the deadline, and even if some machines are in an idle state, the consumed energy is also counted into the total energy consumption.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010315850.6A CN111209104A (en) | 2020-04-21 | 2020-04-21 | Energy perception scheduling method for Spark application under heterogeneous cluster |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010315850.6A CN111209104A (en) | 2020-04-21 | 2020-04-21 | Energy perception scheduling method for Spark application under heterogeneous cluster |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111209104A true CN111209104A (en) | 2020-05-29 |
Family
ID=70787293
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010315850.6A Pending CN111209104A (en) | 2020-04-21 | 2020-04-21 | Energy perception scheduling method for Spark application under heterogeneous cluster |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111209104A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111522647A (en) * | 2020-07-01 | 2020-08-11 | 金陵科技学院 | Public cloud service leasing method capable of minimizing leasing cost |
| CN111736959A (en) * | 2020-07-16 | 2020-10-02 | 南京南软科技有限公司 | Spark task scheduling method considering data affinity under heterogeneous cluster |
| CN113535356A (en) * | 2021-07-01 | 2021-10-22 | 中国科学院软件研究所 | Energy-aware hierarchical task scheduling method and device |
| US11778045B2 (en) | 2021-07-12 | 2023-10-03 | Red Hat, Inc. | Communication system for micro-frontends of a web application |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180143858A1 (en) * | 2016-11-18 | 2018-05-24 | Sam Sanjabi | Method & system for meeting multiple slas with partial qos control |
| CN109241193A (en) * | 2018-09-26 | 2019-01-18 | 联想(北京)有限公司 | The treating method and apparatus and server cluster of distributed data base |
-
2020
- 2020-04-21 CN CN202010315850.6A patent/CN111209104A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180143858A1 (en) * | 2016-11-18 | 2018-05-24 | Sam Sanjabi | Method & system for meeting multiple slas with partial qos control |
| CN109241193A (en) * | 2018-09-26 | 2019-01-18 | 联想(北京)有限公司 | The treating method and apparatus and server cluster of distributed data base |
Non-Patent Citations (2)
| Title |
|---|
| 樊森: "基于异构Spark集群下的Task调度优化方法", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 虞威: "基于能量感知的多数据中心工作流调度方法", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111522647A (en) * | 2020-07-01 | 2020-08-11 | 金陵科技学院 | Public cloud service leasing method capable of minimizing leasing cost |
| CN111522647B (en) * | 2020-07-01 | 2020-10-27 | 金陵科技学院 | Public cloud service leasing method capable of minimizing leasing cost |
| CN111736959A (en) * | 2020-07-16 | 2020-10-02 | 南京南软科技有限公司 | Spark task scheduling method considering data affinity under heterogeneous cluster |
| CN111736959B (en) * | 2020-07-16 | 2020-11-27 | 南京南软科技有限公司 | Spark task scheduling method considering data affinity under heterogeneous cluster |
| CN113535356A (en) * | 2021-07-01 | 2021-10-22 | 中国科学院软件研究所 | Energy-aware hierarchical task scheduling method and device |
| CN113535356B (en) * | 2021-07-01 | 2023-09-12 | 中国科学院软件研究所 | Energy-aware hierarchical task scheduling method and device |
| US11778045B2 (en) | 2021-07-12 | 2023-10-03 | Red Hat, Inc. | Communication system for micro-frontends of a web application |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111209104A (en) | Energy perception scheduling method for Spark application under heterogeneous cluster | |
| Zhu et al. | Scheduling stochastic multi-stage jobs to elastic hybrid cloud resources | |
| US9218213B2 (en) | Dynamic placement of heterogeneous workloads | |
| Żotkiewicz et al. | Minimum dependencies energy-efficient scheduling in data centers | |
| CN109582448B (en) | Criticality and timeliness oriented edge calculation task scheduling method | |
| Saif et al. | Multi-objective grey wolf optimizer algorithm for task scheduling in cloud-fog computing | |
| Nabi et al. | DRALBA: Dynamic and resource aware load balanced scheduling approach for cloud computing | |
| Wu et al. | Optimizing the performance of big data workflows in multi-cloud environments under budget constraint | |
| Liu et al. | A deadline-constrained multi-objective task scheduling algorithm in mobile cloud environments | |
| CN107341041B (en) | Cloud task multidimensional constraint backfill scheduling method based on priority queue | |
| Thaman et al. | Green cloud environment by using robust planning algorithm | |
| Li et al. | Fast and energy-aware resource provisioning and task scheduling for cloud systems | |
| Qu et al. | Study QoS optimization and energy saving techniques in cloud, fog, edge, and IoT | |
| Mishra et al. | Allocation of energy-efficient task in cloud using DVFS | |
| CN106407007B (en) | Cloud resource configuration optimization method for elastic analysis process | |
| Jagadish Kumar et al. | Hybrid gradient descent golden eagle optimization (HGDGEO) algorithm-based efficient heterogeneous resource scheduling for big data processing on clouds | |
| CN108762899B (en) | Cloud task rescheduling method and device | |
| Wang et al. | Exploiting dark cores for performance optimization via patterning for many-core chips in the dark silicon era | |
| Dubey et al. | QoS driven task scheduling in cloud computing | |
| Nicodemus et al. | Managing vertical memory elasticity in containers | |
| CN115562841B (en) | Cloud video service self-adaptive resource scheduling system and method | |
| CN108958919B (en) | Multi-DAG task scheduling cost fairness evaluation method with deadline constraint in cloud computing | |
| Li et al. | PAS: Performance-Aware Job Scheduling for Big Data Processing Systems | |
| CN110308991B (en) | Data center energy-saving optimization method and system based on random tasks | |
| Sun et al. | HEFT-dynamic scheduling algorithm in workflow scheduling |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200529 |