Overview of an SLI analysis

Reading time: 0 minute(s) (0 words)

Besides computing your actual SLO status with the configured metric and time window settings, SLI Analyzer provides you with additional statistical data. These numbers can help you better understand the performance of your service:

Image 1: Ready SLI analysis

Besides the above-mentioned statistical data, you can also see percentiles. These include:

• For the Values field operators ≥ or > , p1, p5, p10, and p50 are displayed:

Image 2: Aggregation for gt and gte operators

• For the Values field operators ≤ or <, p50, p90, p95, p99 and p99.9 are displayed:

Image 3: Aggregation for the lt and lte operators

note

The reliability burn down chart is available and visible only for a successfully performed analysis.

caution

Keep in mind the following assumptions:

• These variables are defined below in terms of time series distributions. A time series distribution is a set of data defined by time: value pairs. The statistical analyses performed by the SLI Analyzer are being performed against the returned data as a distribution as a whole. 

• The values you see for these variables are not directly related to setting up a new SLO. Some are more useful for simply understanding the performance of an SLI, and you can use them to understand the historical performance of your SLI metrics (e.g., during an incident, etc.) or for setting up a new SLO.

SLI charts aggregations

There are different aggregations used for displaying the SLI time series chart.

Threshold metrics

Aggregations in the SLI charts for the threshold metrics follow the below assumptions:

• 95th percentile for operators < or ≤:

Image 4: Aggregation for 95th percentile

• 5th percentile for operators > or ≥

Image 5: Aggregation for 5th percentile

Ratio metrics

Aggregations in the SLI charts for the ratio metrics follow the below assumptions:

• The last value for a given aggregation period for the incremental metrics
• The sum for non-incremental metrics

Image 6: Aggregation for ratio metrics

Definitions of the statistical data in SLI Analyzer

Below, you can find definitions for all statistical data shown in the SLI Analyzer:

• Min: The lowest value in the distribution of the returned SLI values.
• Mean: The average value of the distribution. An average is a value you get when you add up all values in a distribution and then divide that by the total number of data points in the distribution.
• Max: The highest value in the distribution.
• StdDev: The standard deviation of the distribution. A low value of StdDev means that most values are close to the mean. A high value of StdDev means that values vary much more.
• Variance: The number from which a standard deviation is derived. A standard deviation is the square root of variance.
• A variance value tells you how dispersed the numbers in an entire distribution are, while a StdDev gives you a value to set up steps away from the mean.
• Like a standard deviation, a low value means that there is less variability across the distribution, while a high value means there is more.
• Range: (max-min): The delta between the largest and smallest values in the distribution.

Percentiles vs. percentages

It is also essential to understand the difference between a percentile and a percentage.

• A percentage is a measure in relation to 100.  If you have 4 out of 5 of something, you have 80% of that something.

• A percentile is used to compare individual data points in a distribution. If the value of the 90th percentile is 250, you know that 90% of all other data points in the distribution fall below 250.

SLI values distribution charts

An SLI values distribution chart shows the frequency distribution of the data points by collecting them into “buckets” of values that fall within specific ranges. For example, you could say, “We have 5 data points between 0-10, 12 data points between 10-20, and 3 data points between 20-30,” etc.

Nobl9 displays an SLI values distribution chart in the SLI Analyzer as it is much simpler to think about all of the values in a time series distribution at once than it is to try to see how a time series rises and falls over time.

For example, the following is the SLI values distribution chart for a threshold metric:

Image 7: SLI values distribution chart for a threshold metric

The following shows an example of an SLI values distribution chart for a ratio metric:

Image 8: SLI values distribution chart for a ratio metric

Linear and logarithmic scales

You can select a linear or logarithmic scale for the Y-axis in the SLI values distribution chart. Logarithmic scale is useful when its buckets contain a wide range of values:

Linear scale displaying a wide range of values of an SLI metric:

Image 9: SLI values distribution chart for a ratio metric - linear scale

Logarithmic scale offering a more useful insight into a wide range of values of the same metric:

Image 10: SLI values distribution chart for a ratio metric - logarithmic scale

note

According to a convention assumed in SLI Analyzer, the ranges of the buckets in the SLI values distribution chart are shifted. Effectively, the first bucket begins before the minimum value in the distribution of the analyzed time series:

Useful links

What reliability target should I choose?

Choose reliability targets and create meaningful SLOs