WO2025124271A1 - 一种自动修正的算力度量方法、系统及装置 - Google Patents

一种自动修正的算力度量方法、系统及装置 Download PDF

Info

Publication number
WO2025124271A1
WO2025124271A1 PCT/CN2024/137081 CN2024137081W WO2025124271A1 WO 2025124271 A1 WO2025124271 A1 WO 2025124271A1 CN 2024137081 W CN2024137081 W CN 2024137081W WO 2025124271 A1 WO2025124271 A1 WO 2025124271A1
Authority
WO
WIPO (PCT)
Prior art keywords
computing power
power measurement
time
busy
idle
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
Application number
PCT/CN2024/137081
Other languages
English (en)
French (fr)
Inventor
曹俊
杨锦
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.)
China Telecom Cloud Technology Co Ltd
Original Assignee
China Telecom Cloud Technology Co Ltd
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 China Telecom Cloud Technology Co Ltd filed Critical China Telecom Cloud Technology Co Ltd
Publication of WO2025124271A1 publication Critical patent/WO2025124271A1/zh
Anticipated expiration legal-status Critical
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3409—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment for performance assessment
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3051—Monitoring arrangements for monitoring the configuration of the computing system or of the computing system component, e.g. monitoring the presence of processing resources, peripherals, I/O links, software programs
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3452—Performance evaluation by statistical analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G06F11/3476—Data logging
    • 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

Definitions

  • the present invention relates to the field of computing power network technology, and in particular to a method, system and device for automatically correcting computing power measurement.
  • Computing power measurement usually uses the benchmark method to run a load test on a computing power resource of specified specifications, or run a load test periodically to obtain one or more indicators from the application perspective as computing power measurement.
  • the computing power resources provided by data centers to applications are usually virtual machines or containers of certain specifications, which are generated after virtualization and over-allocation on top of physical resources.
  • the actual computing power provided will change over time, and there will be a large difference between idle and busy times of the system.
  • the current computing power measurement method cannot reflect the change of computing power over time, and there are problems such as inaccurate measurement results and excessive resource consumption.
  • One-time test The measurement results are inaccurate, depending on the system busyness and over-allocation strategy during the test, and cannot reflect changes over time.
  • Periodic test The measurement results are inaccurate and consume too much computing power resources. Assuming that the test is performed once a day in the morning, but the system is idle in the morning and busy in the afternoon, the measurement results obtained cannot reflect the situation during the busy hours. Change it to once an hour and then take the average value. The measurement results still cannot reflect changes in the time dimension and consume too much computing power resources.
  • the current computing power measurement method is generally based on the assumption that computing power is continuously stable, which cannot reflect the changes in computing power over time. There are problems such as inaccurate measurement results and excessive resource consumption.
  • the present invention provides an automatically corrected computing power measurement method, system and device, which can output computing power measurement results related to the time dimension and automatically correct the measurement results when the system changes without increasing excessive resource consumption.
  • an automatic correction computing power measurement method comprising:
  • Monitor the performance data of the application analyze the resource consumption index of the unit business volume in the time dimension, determine the low time period of the index value as the idle time, and determine the peak time period of the index value as the busy time;
  • the measurement task manager is re-triggered for testing and a new computing power measurement value is output.
  • CPU utilization and GPU utilization of computing resources are collected, the business concurrency data of the application is collected, and reported to the monitoring data management module.
  • the resource consumption indicator per unit business volume includes: CPU consumption per 100 business concurrency and/or the increase in CPU consumption per 100 additional business concurrency and/or GPU consumption per 100 business concurrency and/or the increase in GPU consumption per 100 additional business concurrency.
  • the monitoring Agent module collects the CPU utilization and GPU utilization of computing resources, collects the application's business concurrency data, and reports it to the monitoring data management module.
  • the method further includes: receiving monitoring data reported by the monitoring Agent module, and saving the monitoring data to a database using the generation time of the monitoring data as an index.
  • the resource consumption index per unit business volume can also be obtained from the computing power resource base.
  • the resource consumption index per unit business volume in the calculation time dimension of the cycle is saved in the database with the generation time of the monitoring data as the index.
  • business concurrency data can also be obtained from the application's operation and maintenance system.
  • the automatic correction method for calculating the computing power of the embodiment of the present invention has the following beneficial effects:
  • Output time dimension computing power measurement results including system idle period, busy period, idle time measurement value, busy time measurement value, more accurately reflect the change of computing power over time; judge the changes of the system by monitoring application performance data, automatically trigger re-measurement, output the corrected computing power measurement value, improve the accuracy of measurement results, and avoid excessive resource consumption caused by periodic testing.
  • the present invention also provides an automatically corrected arithmetic power measurement system, which applies the above-mentioned automatically corrected arithmetic power measurement method, including:
  • the computing resource base manages physical machine clusters and provides virtual machines and containers of various specifications based on virtualization technology
  • Test computing resources virtual machines or containers of specified specifications created on demand, for computing power measurement test tasks
  • Application computing resources are virtual machines or containers of specified specifications used for long-term use, in which applications are running;
  • Application programs which are used to perform actual business, include websites, databases, and game service programs;
  • Benchmark programs typical applications, including: database and Web server programs;
  • the measurement task manager is used to manage the execution process of computing power measurement tasks, including the steps of creating computing power resources to be tested, deploying benchmark programs, executing tests, and collecting test results;
  • the metric scheduler calculates and monitors the CPU consumption index of unit business volume in the time dimension, which is used to identify the idle and busy times of the system and major changes in system performance, generates computing power measurement tasks according to the strategy, and calls the computing power measurement task manager to execute the tasks;
  • Data manager receives performance data reported by monitoring agent and saves it to time series database
  • the database is a database that supports the time dimension and supports query and analysis statements based on time conditions
  • Monitoring Agent collects performance data of computing resources and reports it to the monitoring data manager
  • the present invention also provides an automatically corrected computing power measurement device, which, when in operation, executes the above-mentioned automatically corrected computing power measurement method.
  • FIG. 2 is a statistical diagram of CPU consumption indicators per unit of business volume per day according to an embodiment of the present invention
  • FIG. 4 is a comparison chart of CPU consumption indicators per unit of business volume according to an embodiment of the present invention.
  • first and second are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as “first” or “second” may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, “plurality” means two or more.
  • the terms “installed”, “connected”, and “connected” should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components.
  • installed should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components.
  • the preferred embodiment of the present invention provides an automatically corrected computing power measurement method and device, and sets a computing power measurement scheduler module, the main functions of which include:
  • A. Identification mechanism for busy and idle hours monitor the performance data of the application and analyze the resource consumption indicators per unit of business volume in the time dimension (such as CPU consumption per 100 business concurrency, or the increase in CPU consumption per 100 business concurrency, GPU consumption per 100 business concurrency, or the increase in GPU consumption per 100 business concurrency).
  • the low time period of the indicator value is the idle time
  • the peak time period of the indicator value is the busy time.
  • Time dimension computing power measurement Select one idle time and one busy time of the day/week, trigger the measurement task manager to perform a test respectively, and output the idle period, busy period, idle time measurement value, and busy time measurement value of the system every day/week.
  • C. Automatically correct computing power measurement Continuously monitor the performance data of the application. When the resource consumption index value of the unit business volume in the same period of the previous and next two weeks is found to have a significant change (such as the difference is greater than 30%), re-trigger the measurement task manager for testing and output a new computing power measurement value.
  • Added data manager module the main functions include:
  • the resource consumption index per unit of business volume in the time dimension is calculated periodically (such as every 5 minutes) and saved in the database with the generation time of the monitoring data as the index.
  • VM/container add a monitoring Agent module to collect the CPU utilization and GPU utilization of computing resources (VM/container), collect the application's business concurrency data, and report it to the monitoring data management module.
  • Example 1 Collecting application performance data
  • the application calls the reporting SDK interface to report real-time business concurrency data, such as 100 concurrency/second at 2023-04-16 10:10:10;
  • the monitoring agent receives the reported business concurrency data
  • the monitoring agent collects real-time performance data of the computing resources that have been used, including CPU utilization, GPU utilization, etc., such as 2023-04-16 10:10:10 CPU 80%, 2023-04-16 10:10:10 GPU 28%;
  • the monitoring agent reports real-time performance data and real-time business concurrency data of computing resources
  • the data manager receives the real-time performance data and real-time business concurrency data of computing resources, and saves the data to the time series database according to the time dimension, as shown in Table 1.
  • Example 2 Generating computing power measurement results in the time dimension
  • the processing capacity of the system in different time periods can be measured more accurately, so as to better understand and monitor the system performance.
  • the busy and idle periods of the system in different time periods can be identified, which is of great significance for optimizing resource allocation and improving system efficiency and response speed.
  • the unit business volume resource consumption index counted by day and week the changes in the resource demand of the system in different time periods can be discovered in time, so as to adjust the resource allocation in time to meet the system requirements and improve the system performance and response speed.
  • the output system data such as the high probability idle period, high probability busy period, idle time measurement value, busy time measurement value by day, and the high probability idle period, high probability busy period, idle time measurement value, busy time measurement value by week can provide powerful decision-making support for system administrators and decision makers, helping them to better understand the system operation status, optimize resource allocation, and improve system performance.
  • this embodiment if the number of major change periods exceeds 12, it is considered that the system performance has changed significantly. However, it should be noted that this number is a preset warning value and can be adjusted according to experience and actual conditions;
  • Output new measurement results including the high-probability idle time period, high-probability busy time period, idle time measurement value, and busy time measurement value by day, and the high-probability idle time period, high-probability busy time period, idle time measurement value, and busy time measurement value by week.
  • this embodiment identifies the major change period by comparing the average value of the unit business volume resource consumption index with the data of the same time period last week, so as to monitor whether the system performance has undergone major changes. If the major change period exceeds the preset warning value (for example, 12), it is considered that the system performance has undergone major changes.
  • the preset warning value for example, 12
  • Adjusting resource allocation After identifying significant changes in system performance, this embodiment recalculates the idle and busy hours on a daily and weekly basis, which helps to more accurately identify the busy and idle periods of the system in a day or week, thereby more effectively adjusting resource allocation and improving resource utilization.
  • this embodiment re-calls the measurement task manager for testing during the idle time and busy time on a daily and weekly basis, thereby obtaining more accurate measurement results.
  • the method of automatically correcting measurement results can timely discover and correct possible measurement errors, thereby improving the accuracy and reliability of the measurement.
  • the main function of the measurement task manager is to measure and monitor system performance. It can count the average value of resource consumption indicators per unit of business volume, and compare the data of the same time period last week to identify the period of major changes. In addition, the measurement task manager can also re-count the idle and busy times by day and week to more accurately identify the busy and idle periods of the system in a day or a week, so as to more effectively adjust resource allocation and improve resource utilization. In terms of outputting new measurement results, the measurement task manager re-calls and tests the idle and busy times by day and the idle and busy times by week to obtain more accurate measurement results. These new measurement results, including the high probability idle period, high probability busy period, idle time measurement value and busy time measurement value by day and week, can provide powerful decision support for system administrators and decision makers, helping them to better understand the operation status of the system, optimize resource allocation and improve system performance.
  • the embodiments of the present invention provide an automatically corrected computing power measurement method, system and device. After its actual application, by monitoring and collecting the actual operation performance data of the application, the idle time and busy time of the system are identified, and the idle time and busy time are selected for testing respectively, and the system high probability idle period, high probability busy period, idle time computing power measurement value, and busy time computing power measurement value are output.
  • the actual operation performance data of the application is monitored to have a significant change, it is judged that the system virtualization and over-allocation strategies have changed, and the test is re-performed to output a new computing power measurement value.
  • the system provided in the above embodiment is only illustrated by the division of the above functional modules.
  • the above functions can be assigned to different functional modules as needed, that is, the modules or steps in the embodiments of the present invention can be decomposed or combined.
  • the modules in the above embodiments can be combined into one module, or further divided into multiple sub-modules to complete all or part of the functions described above.
  • the names of the modules and steps involved in the embodiments of the present invention are only for distinguishing the modules or steps, and are not regarded as improper limitations of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Quality & Reliability (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Hardware Design (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Debugging And Monitoring (AREA)

Abstract

本申请涉及算力网络技术领域，具体涉及一种自动修正的算力度量方法、系统及装置，该方法包括：监控应用的性能数据，分析时间维度上的单位业务量资源消耗指标，将所述指标值的低谷时间段判定为闲时，将指标值的高峰时间段判定为忙时；随机选择每天/每周的1个闲时和1个忙时，分别触发度量任务管理器进行一次测试，输出系统每天/每周的空闲时段、繁忙时段、闲时度量值与忙时度量值；当前后两周的相同时段单位业务量资源消耗指标值差异大于预设警戒值，重新触发度量任务管理器进行测试，输出新的算力度量值。本发明输出时间维度算力度量结果，包含系统空闲时段、繁忙时段、闲时度量值、忙时度量值，更准确的反映算力随时间的变化情况。

Description

一种自动修正的算力度量方法、系统及装置
相关申请
本申请要求2023年12月14日申请的，申请号为2023117182398，名称为“一种自动修正的算力度量方法、系统及装置”的中国专利申请的优先权，在此将其全文引入作为参考。
技术领域
本发明涉及算力网络技术领域，具体而言，涉及一种自动修正的算力度量方法、系统及装置。
背景技术
在算力网络领域，准确的算力度量可以提升系统效率，降低应用成本。算力度量通常使用基准程序法，在指定规格的算力资源上跑一次负载测试，或周期跑负载测试，得到应用角度的一项或多项指标作为算力度量。
随着云计算技术的快速发展，数据中心提供给应用的算力资源通常是某种规格的虚拟机或容器，是在物理资源之上进行虚拟化和超分配后产生的，提供的真实算力会随时间变化，在系统闲时和忙时有较大的差异。当前的算力度量方法不能反映算力随时间的变化情况，存在度量结果不准确、资源消耗过多等问题。
一次测试：度量结果不准确，取决于测试时的系统繁忙程度和超分配策略，也不能反映随时间的变化情况。周期测试：度量结果不准确，而且消耗过多的算力资源。假设每天上午测试一次，但系统每天上午是闲时，下午是忙时，那么得到的度量结果不能反映忙时的情况。修改为每小时一次，然后取平均值，得到的度量结果还是不能反映时间维度上的变化，而且消耗过多的算力资源。
当前算力度量方法一般为基于算力持续稳定的假设，不能反映算力随时间的变化情况，存在度量结果不准确、资源消耗过多等问题。
发明内容
有鉴于此，本发明提供了一种自动修正的算力度量方法、系统及装置，能够输出时间维度相关的算力度量结果，并在系统发生变化时自动修正度量结果，且不增加过多的资源消耗。
一方面，本发明提出了一种自动修正的算力度量方法包括：
监控应用的性能数据，分析时间维度上的单位业务量资源消耗指标，将所述指标值的低谷时间段判定为闲时，将指标值的高峰时间段判定为忙时；
随机选择每天/每周的1个闲时和1个忙时，分别触发度量任务管理器进行一次测试，输出系统每天/每周的空闲时段、繁忙时段、闲时度量值与忙时度量值；
当前后两周的相同时段单位业务量资源消耗指标值差异大于预设警戒值，重新触发度量任务管理器进行测试，输出新的算力度量值。
进一步地，收集算力资源的CPU利用率、GPU利用率，收集应用的业务并发量数据，并上报给监控数据管理模块。
进一步地，所述单位业务量资源消耗指标包括：每100业务并发量的CPU消耗和/或每增加100业务并发量的CPU消耗增加量和/或每100业务并发量的GPU消耗和/或每增加100业务并发量的GPU消耗增加量。
进一步地，还包括：
通过监控Agent模块收集算力资源的CPU利用率、GPU利用率，收集应用的业务并发量数据，并上报给监控数据管理模块。
进一步地，还包括：接收监控Agent模块上报的监控数据，以监控数据产生时间为索引保存到数据库。
进一步地，单位业务量资源消耗指标还可通过从算力资源底座获取。
进一步地，周期的计算时间维度上的单位业务量资源消耗指标，以监控数据产生时间为索引保存到数据库。
进一步地，业务并发量数据通过上报SDK模块的接口进行上报；
或业务并发量数据还可通过从应用的运维系统获取。
本发明实施例一种自动修正的算力度量方法，与现有技术相比，其有益效果在于：
输出时间维度算力度量结果，包含系统空闲时段、繁忙时段、闲时度量值、忙时度量值，更准确的反映算力随时间的变化情况；通过监控应用性能数据来判断系统的变化，自动触发重新度量，输出修正后的算力度量值，提升度量结果的准确性，且避免周期测试带来的过多资源消耗。
另一方面，本发明还提供了一种自动修正的算力度量系统，应用上述的自动修正的算力度量方法，包括：
算力资源底座，数据中心的IAAS层，管理物理机集群，基于虚拟化技术来提供各种规格的虚拟机和容器；
测试算力资源，按需创建的指定规格虚拟机或容器，用于算力度量的测试任务；
应用算力资源，作为长期使用的指定规格虚拟机或容器，里面运行着应用程序；
应用程序，承担实际业务的应用程序，包括：网站、数据库、游戏服务程序；
基准程序，典型的应用程序，包括：数据库与Web服务器程序；
度量任务管理器，用于管理算力度量任务的执行过程，包括创建待测试算力资源、部署基准程序、执行测试与收集测试结果步骤；
度量调度器，计算和监控时间维度上的单位业务量CPU消耗指标，用于识别系统闲时、忙时以及系统性能重大变化，按策略生成算力度量任务，并调用算力度量任务管理器执行任务；
数据管理器，接收监控Agent上报的性能数据，并保存到时序数据库；
数据库，为支持时间维度的数据库，支持时间条件的查询和分析语句；
监控Agent，收集算力资源的性能数据，并上报到监控数据管理器；
上传SDK，为应用提供上传业务并发量数据的接口。
另一方面，本发明还提供了一种自动修正的算力度量装置，当其在运行时，执行上述的自动修正的算力度量方法。
可以理解的是，上述自动修正的算力度量系统及装置，与自动修正的算力度量方法具备相同的有益效果，在此不再赘述。
附图说明
为了更清楚地说明本申请实施例或传统技术中的技术方案，下面将对实施例或传统技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据公开的附图获得其他的附图。
图1是本发明实施例一种自动修正的算力度量系统的结构示意图；
图2是本发明实施例单位业务量CPU消耗指标-天的统计图；
图3是本发明实施例单位业务量CPU消耗指标-周的统计图；
图4是本发明实施例单位业务量CPU消耗指标对比图。
具体实施方式
下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
在本申请的描述中，需要理解的是，术语“中心”、“上”、“下”、“前”、“后”、“左”、“右”、“竖直”、“水平”、“顶”、“底”、“内”、“外”等指示的方位或位置关系为基于附图所示的方位或位置关系，仅是为了便于描述本申请和简化描述，而不是指示或暗示所指的装置或元件必须具有特定的方位、以特定的方位构造和操作，因此不能理解为对本申请的限制。
术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括一个或者更多个该特征。在本申请的描述中，除非另有说明，“多个”的含义是两个或两个以上。
在本申请的描述中，需要说明的是，除非另有明确的规定和限定，术语“安装”、“相连”、“连接”应做广义理解，例如，可以是固定连接，也可以是可拆卸连接，或一体地连接；可以是机械连接，也可以是电连接；可以是直接相连，也可以通过中间媒介间接相连，可以是两个元件内部的连通。对于本领域的普通技术人员而言，可以具体情况理解上述术语在本申请中的具体含义。
本发明优选实施例提供了一种自动修正的算力度量方法及装置，并设置算力度量调度器模块，模块的主要功能包括：
A.闲忙时识别机制：监控应用的性能数据，分析时间维度上的单位业务量资源消耗指标(如每100业务并发量的CPU消耗，或每增加100业务并发量的CPU消耗增加量、每100业务并发量的GPU消耗，或每增加100业务并发量的GPU消耗增加量)，指标值的低谷时间段为闲时，指标值的高峰时间段为忙时。
B.时间维度算力度量：选择天/周的1个闲时和1个忙时，分别触发度量任务管理器进行一次测试，输出系统每天/每周的空闲时段、繁忙时段、闲时度量值、忙时度量值。
C.自动修正算力度量：持续监控应用的性能数据，当前后两周的相同时段单位业务量资源消耗指标值发现生大变化时(如差异大于30％)，重新触发度量任务管理器进行测试，输出新的算力度量值。
2.增加数据管理器模块，主要功能包括：
接收监控Agent上报的监控数据，以监控数据产生时间为索引保存到数据库。
周期(如每5分钟)计算时间维度上的单位业务量资源消耗指标，以监控数据产生时间为索引保存到数据库。
3.增加监控Agent模块，收集算力资源(VM/容器)的CPU利用率、GPU利用率，收集应用的业务并发量数据，并上报给监控数据管理模块。
4.增加上报SDK模块，为应用提供上报业务并发量数据的接口。
参阅图1-4所示，下面将列举本发明下的部分实施例和测定数据以阐述本发明的显著的进步：
实施例1：收集应用性能数据
1.应用程序调用上报SDK的接口，上报实时业务并发量数据，如2023-04-1610:10:10为100并发/秒；
2.监控Agent接收上报的业务并发量数据；
3.监控Agent收集已经使用算力资源的实时性能数据，包括CPU利用率、GPU利用率等，如2023-04-1610:10:10CPU 80％，2023-04-1610:10:10GPU 28％；
4.监控Agent上报算力资源的实时性能数据和实时业务并发量数据；
5.数据管理器接收算力资源的实时性能数据和实时业务并发量数据，按照时间维度保存数据到时序数据库，如表1所示。
表1
实施例2：产生时间维度的算力度量结果
1.计算时间维度上的单位业务量资源消耗指标(如每100业务并发量的CPU消耗，或每增加100业务并发量的CPU消耗增加量、每100业务并发量的GPU消耗，或每增加100业务并发量的GPU消耗增加量)；
2.按天的每个小时统计单位业务量资源消耗指标均值的高峰值时间段和低谷值时间段，作为按天的大概率繁忙时段和大概率空闲时段。如图2所示，1:00-14:00是闲时，20:00-23:00是忙时；
3.按周的每个小时统计单位业务量资源消耗指标均值的高峰值时间段和低谷值时间段，作为按周的大概率繁忙时段和大概率空闲时段。如图3所示，周二1:00-14:00是闲时，周六20:00-23:00是忙时；
4.在按天的闲时和忙时分别调用度量任务管理器进行一次测试。在按周的闲时和忙时分别调用度量任务管理器进行一次测试；
5.输出系统按天的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值。输出系统按周的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值。
可以理解的是，通过计算时间维度上的单位业务量资源消耗指标，如每100业务并发量的CPU消耗或每增加100业务并发量的CPU消耗增加量等，可以更准确地衡量系统在不同时间段的处理能力，以便更好地了解和监控系统性能。通过按天和按周统计单位业务量资源消耗指标均值的高峰值时间段和低谷值时间段，可以识别出系统在不同时间段的繁忙和空闲时段，这对于优化资源分配、提高系统效率和响应速度具有重要意义。根据按天和按周统计的单位业务量资源消耗指标，可以及时发现系统在不同时间段的资源需求变化，从而及时调整资源分配，以满足系统的需求，提高系统性能和响应速度。通过在按天的闲时和忙时以及按周的闲时和忙时分别调用度量任务管理器进行测试，可以获取更准确、可靠的度量结果，从而更好地了解和监控系统性能，为优化资源分配提供更准确的决策支持。输出系统按天的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值以及按周的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值等数据，可以为系统管理员和决策者提供有力的决策支持，帮助他们更好地理解系统的运行状况，优化资源分配，提高系统性能。
实施例3：自动修正算力度量结果
1.按天的每个小时统计单位业务量资源消耗指标均值，与上周相同时段的统计数据进行对比，统计变化幅度超过30％的时段作为重大变化时段。如果重大变化时段超过12个，则认为系统性能发生重大变化，如图4所示；
本实施例优选为重大变化时段超过12个，则认为系统性能发生重大变化。但需要说明的是，该数量作为预设警戒值，可以根据经验与实际情况进行调整；
2.重新统计按天和按周的闲时和忙时；
3.在按天的闲时和忙时分别重新调用度量任务管理器进行一次测试。在按周的闲时和忙时分别重新调用度量任务管理器进行一次测试。
4.输出新的度量结果，包括按天的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值，包括按周的大概率空闲时段、大概率繁忙时段、闲时度量值、忙时度量值。
需要说明的是，监控系统性能方面：该实施例通过对比单位业务量资源消耗指标均值与上周相同时间段的数据，识别出重大变化时段，从而监控系统性能是否发生重大变化。如果重大变化时段超过预设警戒值(例如12个)，则认为系统性能发生重大变化。
调整资源分配：在识别出系统性能的重大变化后，该实施例重新统计了按天和按周的闲时和忙时，有助于更准确地识别系统在一天或一周中的忙闲时段，从而可以更有效地调整资源分配，提高资源利用率。
自动修正度量结果方面：本实施例在重新统计了闲时和忙时后，分别在按天和按周的闲时和忙时重新调用了度量任务管理器进行测试，从而获取更准确的度量结果。自动修正度量结果的方法可以及时发现并纠正可能的度量错误，提高度量的准确性和可靠性。
度量任务管理器的作用主要是进行系统性能的度量和监控。它能够统计单位业务量资源消耗指标均值，并对比与上周相同时间段的数据，识别出重大变化时段。此外，度量任务管理器还可以重新统计按天和按周的闲时和忙时，以便更准确地识别系统在一天或一周中的忙闲时段，从而可以更有效地调整资源分配，提高资源利用率。在输出新的度量结果方面，度量任务管理器在按天的闲时和忙时以及按周的闲时和忙时分别重新调用进行测试，以获取更准确的度量结果。这些新的度量结果，包括按天和按周的大概率空闲时段、大概率繁忙时段、闲时度量值和忙时度量值，可以为系统管理员和决策者提供有力的决策支持，帮助他们更好地理解系统的运行状况，优化资源分配，提高系统性能。
综上，本发明实施例提供一种自动修正的算力度量方法、系统及装置，其实际应用后，通过监控和收集应用的实际运行性能数据，识别系统的闲时和忙时，选择闲时和忙时分别进行一次测试，输出系统大概率空闲时段、大概率繁忙时段、闲时算力度量值、忙时算力度量值。当监控到应用的实际运行性能数据发生重大变化时，判断系统虚拟化和超分配的策略已经发生变化，重新进行测试，输出新的算力度量值。
以上所述仅为本发明的一个实施例子，但不能以此限制本发明的范围，凡依据本发明所做的结构上的变化，只要不失本发明的要义所在，都应视为落入本发明保护范围之内受到制约。
需要说明的是，上述实施例提供的系统，仅以上述各功能模块的划分进行举例说明，在实际应用中，可以根据需要而将上述功能分配由不同的功能模块来完成，即将本发明实施例中的模块或者步骤再分解或者组合，例如，上述实施例的模块可以合并为一个模块，也可以进一步拆分成多个子模块，以完成以上描述的全部或者部分功能。对于本发明实施例中涉及的模块、步骤的名称，仅仅是为了区分各个模块或者步骤，不视为对本发明的不当限定。
术语“包括”或者任何其它类似用语旨在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备/装置不仅包括那些要素，而且还包括没有明确列出的其它要素，或者还包括这些过程、方法、物品或者设备/装置所固有的要素。
至此，已经结合附图所示的优选实施方式描述了本发明的技术方案，但是，本领域技术人员容易理解的是，本发明的保护范围显然不局限于这些具体实施方式。在不偏离本发明的原理的前提下，本领域技术人员可以对相关技术特征作出等同的更改或替换，这些更改或替换之后的技术方案都将落入本发明的保护范围之内。
以上所述，仅为本发明的较佳实施例而已，并非用于限定本发明的保护范围。
以上所述实施例的各技术特征可以进行任意的组合，为使描述简洁，未对上述实施例中的各个技术特征所有可能的组合都进行描述，然而，只要这些技术特征的组合不存在矛盾，都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式，其描述较为具体和详细，但并不能因此而理解为对申请专利范围的限制。应当指出的是，对于本领域的普通技术人员来说，在不脱离本申请构思的前提下，还可以做出若干变形和改进，这些都属于本申请的保护范围。因此，本申请专利的保护范围应以所附权利要求为准。

Claims (10)

  1. 一种自动修正的算力度量方法，其特征在于，包括：
    监控应用的性能数据，分析时间维度上的单位业务量资源消耗指标，将所述指标值的低谷时间段判定为闲时，将指标值的高峰时间段判定为忙时；
    随机选择每天/每周的1个闲时和1个忙时，分别触发度量任务管理器进行一次测试，输出系统每天/每周的空闲时段、繁忙时段、闲时度量值与忙时度量值；
    当前后两周的相同时段单位业务量资源消耗指标值差异大于预设警戒值，重新触发度量任务管理器进行测试，输出新的算力度量值。
  2. 根据权利要求1所述的自动修正的算力度量方法，其特征在于，
    收集算力资源的CPU利用率、GPU利用率，收集应用的业务并发量数据，并上报给监控数据管理模块。
  3. 根据权利要求1所述的自动修正的算力度量方法，其特征在于，所述单位业务量资源消耗指标包括：每100业务并发量的CPU消耗和/或每增加100业务并发量的CPU消耗增加量和/或每100业务并发量的GPU消耗和/或每增加100业务并发量的GPU消耗增加量。
  4. 根据权利要求1所述的自动修正的算力度量方法，其特征在于，还包括：
    通过监控Agent模块收集算力资源的CPU利用率、GPU利用率，收集应用的业务并发量数据，并上报给监控数据管理模块。
  5. 根据权利要求4所述的自动修正的算力度量方法，其特征在于，还包括：接收监控Agent模块上报的监控数据，以监控数据产生时间为索引保存到数据库。
  6. 根据权利要求4所述的自动修正的算力度量方法，其特征在于，单位业务量资源消耗指标还可通过从算力资源底座获取。
  7. 根据权利要求5所述的自动修正的算力度量方法，其特征在于，
    周期的计算时间维度上的单位业务量资源消耗指标，以监控数据产生时间为索引保存到数据库。
  8. 根据权利要求5所述的自动修正的算力度量方法，其特征在于，业务并发量数据通过上报SDK模块的接口进行上报；
    或业务并发量数据通过从应用的运维系统获取。
  9. 一种自动修正的算力度量系统，应用于如权利要求1-8任一项所述的自动修正的算力度量方法中，其特征在于，包括：
    算力资源底座，数据中心的IAAS层，管理物理机集群，基于虚拟化技术来提供各种规格的虚拟机和容器；
    测试算力资源，按需创建的指定规格虚拟机或容器，用于算力度量的测试任务；
    应用算力资源，作为长期使用的指定规格虚拟机或容器，里面运行着应用程序；
    应用程序，承担实际业务的应用程序，包括：网站、数据库、游戏服务程序；
    基准程序，典型的应用程序，包括：数据库与Web服务器程序；
    度量任务管理器，用于管理算力度量任务的执行过程，包括创建待测试算力资源、部署基准程序、执行测试与收集测试结果步骤；
    度量调度器，计算和监控时间维度上的单位业务量CPU消耗指标，用于识别系统闲时、忙时以及系统性能重大变化，按策略生成算力度量任务，并调用算力度量任务管理器执行任务；
    数据管理器，接收监控Agent上报的性能数据，并保存到时序数据库；
    数据库，为支持时间维度的数据库，支持时间条件的查询和分析语句；
    监控Agent，收集算力资源的性能数据，并上报到监控数据管理器；
    上传SDK，为应用提供上传业务并发量数据的接口。
  10. 一种自动修正的算力度量装置，其特征在于，当其在运行时，执行如权利要求1-8任一项所述的自动修正的算力度量方法。
PCT/CN2024/137081 2023-12-14 2024-12-05 一种自动修正的算力度量方法、系统及装置 Pending WO2025124271A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202311718239.8 2023-12-14
CN202311718239.8A CN117851201A (zh) 2023-12-14 2023-12-14 一种自动修正的算力度量方法、系统及装置

Publications (1)

Publication Number Publication Date
WO2025124271A1 true WO2025124271A1 (zh) 2025-06-19

Family

ID=90533621

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/137081 Pending WO2025124271A1 (zh) 2023-12-14 2024-12-05 一种自动修正的算力度量方法、系统及装置

Country Status (2)

Country Link
CN (1) CN117851201A (zh)
WO (1) WO2025124271A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120704615A (zh) * 2025-08-25 2025-09-26 浪潮电子信息产业股份有限公司 一种存储系统的垃圾回收方法和电子设备

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117851201A (zh) * 2023-12-14 2024-04-09 天翼云科技有限公司 一种自动修正的算力度量方法、系统及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120271670A1 (en) * 2011-04-21 2012-10-25 Efficiency3 Corp. Methods, technology, and systems for quickly enhancing the operating and financial performance of energy systems at large facilities, interpreting usual and unusual patterns in energy consumption, identifying, quantifying, and monetizing hidden operating and financial waste, and accurately measuring the results of implemented energy management solutions, in the shortest amount of time with minimal cost and effort
CN110995614A (zh) * 2019-11-05 2020-04-10 华为技术有限公司 一种算力资源分配的方法及装置
CN112088365A (zh) * 2018-03-23 2020-12-15 卡罗莱纳云交换公司 将不同计算资源的使用量化为单个测量单元
CN115334560A (zh) * 2022-08-19 2022-11-11 中国联合网络通信集团有限公司 基站异常的监测方法、装置、设备及计算机可读存储介质
CN117851201A (zh) * 2023-12-14 2024-04-09 天翼云科技有限公司 一种自动修正的算力度量方法、系统及装置

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8578023B2 (en) * 2011-07-29 2013-11-05 Hewlett-Packard Development Company, L.P. Computer resource utilization modeling for multiple workloads
WO2016068930A1 (en) * 2014-10-30 2016-05-06 Hewlett Packard Enterprise Development Lp Resource consuming tasks scheduler
CN112667392B (zh) * 2020-12-09 2024-01-23 南方电网数字电网研究院有限公司 云计算资源分配方法、装置、计算机设备和存储介质
CN113568822B (zh) * 2021-08-03 2023-09-05 安天科技集团股份有限公司 业务资源监控方法、装置、计算设备及存储介质
CN113656174A (zh) * 2021-08-18 2021-11-16 河北幸福消费金融股份有限公司 资源分配方法、系统、计算机设备和存储介质
CN116360972A (zh) * 2021-12-27 2023-06-30 华为技术有限公司 资源管理方法、装置及资源管理平台
CN114756340A (zh) * 2022-03-17 2022-07-15 中国联合网络通信集团有限公司 算力调度系统、方法、装置和存储介质
CN116418876A (zh) * 2023-04-27 2023-07-11 杭州东方通信软件技术有限公司 一种算力网络服务的迁移方法、系统及云管理平台
CN116700962A (zh) * 2023-05-30 2023-09-05 浪潮软件科技有限公司 一种基于karmada的异构算力跨云应用编排方法
CN117170869A (zh) * 2023-09-05 2023-12-05 抖音视界有限公司 数据中心资源消耗量的度量方法、装置、设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120271670A1 (en) * 2011-04-21 2012-10-25 Efficiency3 Corp. Methods, technology, and systems for quickly enhancing the operating and financial performance of energy systems at large facilities, interpreting usual and unusual patterns in energy consumption, identifying, quantifying, and monetizing hidden operating and financial waste, and accurately measuring the results of implemented energy management solutions, in the shortest amount of time with minimal cost and effort
CN112088365A (zh) * 2018-03-23 2020-12-15 卡罗莱纳云交换公司 将不同计算资源的使用量化为单个测量单元
CN110995614A (zh) * 2019-11-05 2020-04-10 华为技术有限公司 一种算力资源分配的方法及装置
CN115334560A (zh) * 2022-08-19 2022-11-11 中国联合网络通信集团有限公司 基站异常的监测方法、装置、设备及计算机可读存储介质
CN117851201A (zh) * 2023-12-14 2024-04-09 天翼云科技有限公司 一种自动修正的算力度量方法、系统及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN120704615A (zh) * 2025-08-25 2025-09-26 浪潮电子信息产业股份有限公司 一种存储系统的垃圾回收方法和电子设备

Also Published As

Publication number Publication date
CN117851201A (zh) 2024-04-09

Similar Documents

Publication Publication Date Title
US12333348B2 (en) Allocating computing resources for deferrable virtual machines
CN111580934B (zh) 云计算环境下多租户虚拟机性能一致的资源分配方法
US20130158950A1 (en) Application performance analysis that is adaptive to business activity patterns
Jassas et al. Failure analysis and characterization of scheduling jobs in google cluster trace
CN103186603B (zh) 确定sql语句对关键业务的性能的影响的方法、系统和设备
CN117851201A (zh) 一种自动修正的算力度量方法、系统及装置
CN107247651A (zh) 云计算平台监测预警方法和系统
Ai et al. On elasticity measurement in cloud computing
CN102298542A (zh) 一种应用程序的质量确定方法和系统
CN107368372A (zh) 一种基于云海os平台的资源展示方法与装置
JP5833125B2 (ja) システムパフォーマンスのリアルタイム測定方法及び装置
CN109525036A (zh) 一种通信设备的市电供应状态的监控方法、装置和系统
CN121411721A (zh) 一种多打印机的自助打印调控方法及系统
CN115640196A (zh) 一种基于链路跟踪的性能预测告警方法及系统
CN111708671A (zh) 一种对服务器资源使用效能评估及资源调配的方法及系统
CN120104440B (zh) 火焰图的生成方法和装置、存储介质及电子设备
CN120631696A (zh) 一种内核进程调度关联信息监测方法、设备、介质及产品
US12554301B2 (en) Estimating power consumption of software workloads
CN118779143A (zh) 一种基于数据分析的算力异常检测方法及系统
Panneerselvam et al. Characterisation of hidden periodicity in large-scale cloud datacentre environments
CN118277103B (zh) 资源调度管理方法、智能计算云操作系统及计算平台
CN118840169B (zh) 一种基于事件的云资源计量方法和系统
CN111475261A (zh) 一种云平台弹性度量方法、系统和存储介质
CN119292760B (zh) 基于分布式多任务请求的云计算系统
CN120560929A (zh) 一种处理器繁忙程度计算方法、装置、车辆及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24902692

Country of ref document: EP

Kind code of ref document: A1