How to read the graph ☝

• This graph shows response time (not execution time) clusters for each platform.
• All services are implemented in Typescript, not all platforms tested support full Typescript or other languages.
• All platforms are triggered from the same machine on a standard Danish consumer network.
• It is split in two, showing cascading cold-start times on the left, and hot-start times on the right. Cold-start is when a service is triggered for the first time in a while. 'Cascading cold-start' means the total time for one cold service to trigger another cold service. Hot-start is when a service is hit repeatedly.
• The graph has two y-axies, purple points use the purple (left) axis, green points use the green (right) axis. Notice that the green points are zoomed in 20x, thus all green points are below the lowest half line of purple columns, ie. much much faster than the purple points.
• If there are one or two points capped at the top they are outliers, and disregarded.
• Points on 0 means the service failed to respond. This also adds a ❗ sign to the platform name.
• Columns (platforms) are sorted by the average response-time (outliers disregarded).
• The graph shows around ~100 data points, so one point corresponds to 1% of calls.
• This graph illustrates the range and spread of response times. You can also gauge a platforms tendency for outliers.

Time series

How to read the graphs 👇

• These graphs show response-times over time, for hot-starts and cold-starts respectively.
• Platforms with zero data points in the plotted interval are grayed out in the bottom.
• The graph also shows a three day running average (outliers removed), used to see if a platforms response times are improving over time.

Debugging

Nevermind this graph ❗