您现在访问的是微软AZURE全球版技术文档网站,若需要访问由世纪互联运营的MICROSOFT AZURE中国区技术文档网站,请访问 https://docs.azure.cn.

解释记分卡Interpreting your scorecard

"记分卡" 选项卡包含测试的聚合结果和分析结果。The scorecard tab contains the aggregated and analyzed results from your tests. 每个测试都有其自己的记分卡。Each test has its own scorecards. 记分卡提供测量结果的快速和有意义的汇总,为网络要求提供数据驱动的结果。Scorecards provide quick and meaningful summaries of measurement results to provide data-driven results for your networking requirements. Internet 分析器负责分析,使你能够专注于决定。Internet Analyzer takes care of the analysis, allowing you to focus on the decision.

"记分卡" 选项卡可在 Internet 分析器资源菜单中找到。The scorecard tab can be found in the Internet Analyzer resource menu.


  • 测试: 选择要查看其结果的测试-每个测试都有其自己的记分卡。Test: Select the test that you’d like to view results for - each test has its own scorecard. 当有足够的数据来完成分析时,将显示测试数据–在大多数情况下,这应在24小时内。Test data will appear once there is enough data to complete the analysis – in most cases, this should be within 24 hours.
  • 时间段 & 结束日期: 每日生成三个记分卡–每个记分卡反映不同的聚合周期–24小时之前的时间(天)、7天前(周)和30天(月)。Time period & end date: Three scorecards are generated daily – each scorecard reflects a different aggregation period – the 24 hours prior (day), the seven days prior (week), and the 30 days prior (month). 使用 "结束日期" 筛选器选择要查看的时间段的最后一天。Use the “End Date” filter to select the last day of the time period you want to see.
  • 国家/地区: 对于您拥有最终用户的每个国家/地区,都将生成一个记分卡。Country: For each country that you have end users, a scorecard is generated. 全局筛选器包含所有最终用户。The global filter contains all end users.

度量值计数Measurement count

度量值的数量会影响分析的置信度。The number of measurements impacts the confidence of the analysis. 计数越高,结果就越精确。The higher the count, the more accurate the result. 对于每个终结点,每个终结点至少应达到100个度量值。At minimum, tests should aim for a minimum of 100 measurements per endpoint per day. 如果度量值计数太小,请将 JavaScript 客户端配置为在应用程序中更频繁地执行。If measurement counts are too low, please configure the JavaScript client to execute more frequently in your application. 对于终结点 A 和 B,它们的度量计数应该非常相似,但这种情况并不常见。The measurement counts for endpoints A and B should be very similar although small differences are expected and okay. 对于较大的差异,不应信任结果。In the case of large differences, the results should not be trusted.


延迟时间(以毫秒为单位)是在 Internet 上的源和目标之间测量速度的常用指标。Latency, measured in milliseconds, is a popular metric for measuring speed between a source and destination on the Internet. 延迟数据通常不会进行分布(即不会跟随 "钟形曲线"),因为在使用统计信息(如算术平均值)时,有 "长尾" 的 "长滞后时间" 值,这些值会使结果倾斜。Latency data is not normally distributed (i.e. does not follow a "Bell Curve") because there is a "long-tail" of large latency values that skew results when using statistics such as the arithmetic mean. 作为替代方法,百分位数提供了一种用于分析数据的 "无分发" 方式。As an alternative, percentiles provide a "distribution-free" way to analyze data. 例如,中值(或第50百分位)总结了分布的中间部分,这些值在其上方,一半位于其下方。As an example, the median, or 50th percentile, summarizes the middle of the distribution - half the values are above it and half are below it. 第75百分位值意味着它大于分布中所有值的75%。A 75th percentile value means it is larger than 75% of all values in the distribution. Internet 分析器将速记中的百分位数称为 P50、P75 和 P95。Internet Analyzer refers to percentiles in shorthand as P50, P75, and P95.

Internet 分析器百分位数是_示例指标_。Internet Analyzer percentiles are sample metrics. 这与真实的_总体指标_不同。This is in contrast to the true population metric. 例如,南大学和 Microsoft 的一名大学的学生之间的每日真实人口中等延迟时间是一天中所有请求的中间延迟值。For example, the daily true population median latency between students at the University of Southern California and Microsoft is the median latency value of all requests during that day. 在实践中,测量所有请求的值是不切实际的,因此我们假定一个合理的示例代表真正的总体。In practice, measuring the value of all requests is impractical, so we assume that a reasonably large sample is representative of the true population.

出于分析目的,P50 (中间值)可用作延迟分布的预期值。For analysis purposes, P50 (median), is useful as an expected value for a latency distribution. 较高的百分位数(例如 P95)可用于识别最差情况下的高延迟。Higher percentiles, such as P95, are useful for identifying how high latency is in the worst cases. 如果你有兴趣了解客户的延迟,一般情况下,P50 是要关注的正确指标。If you are interested in understanding customer latency in general, P50 is the correct metric to focus on. 如果您关心如何了解性能最差的客户,则应关注 P95。If you are concerned with understanding performance for the worst-performing customers, then P95 should be the focus. P75 是这两者之间的平衡点。P75 is a balance between these two.


增量是指终结点 A 和 B 的度量值之间的差异。计算增量的值是为了显示 B 在上的好处。正值表示 B 的执行效果比 A 更好,而负值表示 B 的性能较差。A delta is the difference in metric values for endpoints A and B. Deltas are computed to show the benefit of B over A. Positive values indicate B performed better than A, whereas negative values indicate B's performance is worse. 增量可以是绝对值(例如10毫秒)或相对(5%)。Deltas can be absolute (e.g. 10 milliseconds) or relative (5%).

置信区间Confidence interval

置信区间(CI)是一个值范围,其中包含总体指标,如中间值、P75 或 average。Confidence intervals (CI) are a range of values that have a probability of containing the population metric such as median, P75, or average. 我们遵循使用 95% CI 的常见统计约定。We follow the common statistical convention of using the 95% CI.

对于 Internet 分析器,较窄的置信区间非常好,因为它显示样本指标可能非常接近实际总体指标。For Internet Analyzer, a narrow confidence interval is good because it shows the sample metric is likely very close to the actual population metric. 宽置信时间间隔意味着样本指标反映真实总体指标的确定性更小。A wide confidence interval means less certainty that our sample metric reflects the true population metric. 改善 CI 的最佳方式是增加度量计数。The best way to improve the CI is to increase measurement counts.

时序Time series

时序显示某个指标随时间变化的方式。A time series shows how a metric changes over time. 在 Internet 上,有许多影响性能的时态因素,如高峰流量期、工作日-周末人口差和假期。On the Internet, there are many temporal factors that impact performance such as peak traffic periods, weekday-weekend population differences, and holidays.

后续步骤Next steps

若要了解详细信息,请参阅我们的Internet 分析器概述To learn more, see our Internet Analyzer Overview.