了解功能Understanding capabilities

适用于:Windows Server 2019Applies To: Windows Server 2019

本主题定义了 System Insights 中的功能概念,并介绍了 Windows Server 2019 中提供的默认功能。This topic defines the concept of capabilities in System Insights and introduces the default capabilities available in Windows Server 2019.

本主题还介绍了用于默认功能的数据源、预测时间线和预测状态。This topic also describes the data sources, prediction timelines, and prediction statuses used for the default capabilities.

功能概述Capability overview

系统见解功能是一种机器学习或统计模型,可对系统数据进行分析,以帮助您更深入地了解部署的功能。A System Insights capability is a machine learning or statistics model that analyzes system data to help give you increased insight into the functioning of your deployment. 系统见解引入了一组初始默认功能,它允许您动态添加新功能,而无需更新操作系统。System Insights introduces an initial set of default capabilities, and it allows you to add new capabilities dynamically, without needing to update the operating system.

备注

本文提供了说明如何创建、添加和更新功能的详细文档管理功能文档提供了有关此功能的更多高级信息。Detailed documentation explaining how to create, add, and update capabilities is available here, and the managing capabilities document provides more high-level information about this functionality.

此外,每项功能在 Windows Server 实例上以本地方式运行,并且每个功能都可以单独管理。Additionally, each capability runs locally on a Windows Server instance, and each capability can be managed individually.

功能输出Capability outputs

调用功能时,它将提供一个输出,以帮助解释其分析或预测的结果。When a capability is invoked, it provides an output to help explain the result of its analysis or prediction. 每个输出都必须包含一个状态说明和一个状态说明来描述预测,并且每个结果都可以有选择性地包含与预测关联的特定于功能的数据。Each output must contain a Status and a Status Description to describe the prediction, and each result can optionally contain capability-specific data associated with the prediction. 状态说明可帮助提供状态的上下文说明,并报告 "确定"、"****警告" 或 "严重" 状态。The Status Description helps provides a contextual explanation for the Status, and the capability reports either an OK, Warning, or Critical status. 此外,如果未进行预测,则功能可以使用错误状态。Additionally, a capability can use an Error or None status if no prediction was made. 下面是功能状态及其基本含义:Together, here are the capability statuses and their basic meanings:

  • 确定-一切看起来都不错。Ok - Everything looks good.
  • 警告-无需立即关注,但应查看。Warning - No immediate attention required, but you should take a look.
  • 严重-应立即进行查看。Critical - You should take a look soon.
  • 错误-未知问题导致功能失败。Error - An unknown problem caused the capability to fail.
  • -未进行预测。None - No prediction was made. 这可能是由于缺少数据或任何其他特定于功能的原因而不进行预测。This could be due to a lack of data or any other capability-specific reason for not making a prediction.

此外,结果中包含的任何特定于功能的数据都将放置在用户可访问的 JSON 文件中,并且可以使用 PowerShell 找到文件路径。Additionally, any capability-specific data contained in the result will be placed in a user-accessible JSON file, and the file path can be found using PowerShell.

默认功能Default capabilities

在 Windows Server 2019 中,System Insights 引入了四个侧重于容量预测的默认功能:In Windows Server 2019, System Insights introduces four default capabilities focused on capacity forecasting:

  • Cpu 容量预测-预测 cpu 使用率。CPU capacity forecasting - Forecasts CPU usage.
  • 网络容量预测-预测每个网络适配器的网络使用情况。Networking capacity forecasting - Forecasts network usage for each network adapter.
  • 总存储消耗预测-预测所有本地驱动器上的总存储使用量。Total storage consumption forecasting - Forecasts total storage consumption across all local drives.
  • 卷消耗预测-预测每个卷的存储消耗量。Volume consumption forecasting - Forecasts storage consumption for each volume.

每项功能将分析过去的历史数据以预测未来的使用情况,并且所有预测功能都设计为预测长期趋势而不是短期行为,从而帮助管理员正确预配硬件并调整其工作负荷,以避免将来发生资源争用。Each capability analyzes past historical data to predict future usage, and all of the forecasting capabilities are designed to forecast long-term trends rather than short-term behavior, helping administrators correctly provision hardware and tune their workloads to avoid future resource contention. 由于这些功能侧重于长期使用,因此这些功能分析日常数据。Because these capabilities focus on long-term usage, these capabilities analyze daily data.

预测模型Forecasting model

默认功能使用预测模型来预测将来的使用情况,对于每个预测,该模型将在本地对计算机数据进行训练。The default capabilities use a forecasting model to predict future usage, and for each prediction, the model is trained locally on your machine's data. 此模型旨在帮助检测更长的术语趋势,并在每个 Windows Server 实例上重新训练,使您能够适应每台计算机使用情况的特定行为和细微差别。This model is designed to help detect longer term trends, and retraining on each Windows Server instance enables the capability to adapt to the specific behavior and nuances of each machine's usage.

备注

确定要使用的模型类型需要使用包含数十台计算机的数据集来测试多个模型。Determining what type of model to use required testing many models using a dataset containing tens of thousands of machines. 在分析和调整这些模型后,我们决定使用自动回归预测模型,因为它会产生非常准确且直观直观的预测,而不需要太多的时间进行训练。After analyzing and tweaking these models, we decided to use an auto-regressive forecasting model, as it produces highly-accurate and visually intuitive predictions while not requiring too much time to train. 但是,此模型需要三周的定型数据,因此,每项功能都使用基本的线性趋势,直到有三个周的数据可用。This model, however, requires three weeks of training data, so each capability uses a basic linear trend until three weeks of data are available.

预测时间线Forecasting timelines

默认功能根据收集的数据的天数预测将来特定天数内的时间。The default capabilities forecast a certain number of days into the future based on the number of days for which data has been collected. 下表显示了这些功能的预测时间线:The following table shows the prediction timelines of these capabilities:

输入数据大小Input data size 预测长度Forecast length
0-5 天0-5 days 不进行任何预测。No prediction is made.
6-180 天6-180 days 1/3 * 输入数据的大小1/3 * size of input data
180-365 天180-365 days 60天60 days

预测数据Forecasting data

每个功能分析每日数据以预测未来使用情况。Each capability analyzes daily data to forecast future usage. 但 CPU、网络和存储使用量甚至可以在一天中频繁更改,从而动态调整计算机上的工作负载。CPU, networking, and even storage usage, however, can frequently change throughout the day, dynamically adjusting to the workloads on the machine. 由于在一整天内使用不稳定,因此,在单个数据点中正确地表示每日使用非常重要。Because usage isn't constant throughout the day, it's important to properly represent daily usage in a single data point. 下表详细列出了特定数据点和数据的处理方式:The table below details the specific data points and how the data is processed:

功能名称Capability name 数据源 (s) Data source(s) 筛选逻辑Filtering logic
卷消耗预测Volume consumption forecasting 卷大小Volume size 每日最大使用量Maximum daily usage
总存储消耗预测Total storage consumption forecasting 卷大小和磁盘大小之和Sum of volume sizes, sum of disk sizes 每日最大使用量Maximum daily usage
CPU 容量预测CPU capacity forecasting 处理器时间百分比% Processor Time 每日最大2小时平均Maximum 2-hour average per day
网络容量预测Networking capacity forecasting 字节总数/秒Bytes Total/sec 每日最大2小时平均Maximum 2-hour average per day

在评估上述筛选逻辑时,务必要注意的是,每个功能在未来使用会有意义地超过可用容量时设法通知管理员–即使 CPU 暂时达到了100% 的利用率,CPU 使用率也可能不会导致性能下降或资源争用。When evaluating the filtering logic above, it's important to note that each capability seeks to inform administrators when future usage will meaningfully exceed the available capacity – even though CPU momentarily hit 100% utilization, CPU usage may not have caused meaningful performance degradation or resource contention. 对于 CPU 和网络,应该存在持续的高利用率,而不是瞬时高峰。For CPU and networking, then, there should be sustained high usage rather than momentary spikes. 但是,在整个日子内平均 CPU 和网络使用情况会丢失重要的使用情况信息,因为在几个小时的 CPU 或网络使用情况下,可能会对关键工作负荷的性能产生影响。Averaging CPU and networking usage throughout the whole day, however, would lose important usage information, as a few hours of high CPU or networking usage could meaningfully impact the performance of your critical workloads. 每一天最多2小时的平均平均值将避免这些极端,并为每个要分析的功能生成有意义的数据。The maximum 2-hour average during each day avoids these extremes and still produces meaningful data for each capability to analyze.

不过,对于卷和总存储使用量,存储使用量不能超过可用容量(即使是这样),因此,每日使用的最大值可用于这些功能。For volume and total storage usage, however, storage usage can't exceed the available capacity, even momentarily, so the maximum daily usage is used for these capabilities.

预测状态Forecasting statuses

所有系统见解功能都必须输出与每个预测关联的状态。All System Insights capabilities must output a status associated with each prediction. 每个默认功能使用以下逻辑定义每个预测状态:Each default capability uses the following logic to define each prediction status:

  • 确定:预测不会超过可用容量。OK: The forecast does not exceed the available capacity.
  • 警告:预测超出了未来30天的可用容量。Warning: The forecast exceeds the available capacity in the next 30 days.
  • 严重:预测超出未来7天的可用容量。Critical: The forecast exceeds the available capacity in the next 7 days.
  • 错误:该功能遇到意外错误。Error: The capability ran into an unexpected error.
  • None:没有足够的数据来进行预测。None: There isn't enough data to make a prediction. 这可能是由于缺少数据或者最近未报告任何数据。This could be due to a lack of data or because no data has been reported recently.

备注

如果功能在多个实例上进行预测(例如多个卷或网络适配器),则状态将反映所有实例中的最严重状态。If a capability forecasts on multiple instances - such as multiple volumes or network adapters - the status reflects the most severe status across all instances. 每个卷或网络适配器的各个状态在 Windows 管理中心或每个功能的输出中所包含的数据中可见。Individual statuses for each volume or network adapter are visible in Windows Admin Center or within the data contained in the output of each capability. 有关如何分析默认功能的 JSON 输出的说明,请访问此博客For instructions on how to parse the JSON output of the default capabilities, visit this blog.

其他参考Additional References

若要了解有关系统见解的详细信息,请使用以下资源:To learn more about System Insights, use the following resources: