Benchmarks

[kkaefer]

Add a benchmark mode that allows us to benchmark various aspects of the renderer:

• tile parse time
• stylesheet parse
• stylesheet cascade time
• render time

This requires building uncoupling of these components, and making sure that neither of these actions waits for any I/O, e.g. by implementing a custom mock HTTP object that reads the responses from memory (make sure to preload it from disk).

[kkaefer]

Working on this in https://github.com/mapbox/mapbox-gl-native/tree/benchmarking.

Currently, the only thing this measures is render time. However, render time is highly variable. Often seeing stddev that is higher than the median. However, the median is usually pretty close to the mean, so there are only a handful outliers (typically <0.5% away from 2 standard deviations). If a frame is an outlier, they are typically dramatically outside the mean, typically, by at least a magnitude. The time interval between outliers seems pretty random.

Next actions are:

• find the source of the outliers
• set up a way to track benchmark speeds over time
• get a dedicated benchmark box that is not in the cloud™
• use real hardware, like an ipad/ipod touch/android phones to execute the benchmarks

[jfirebaugh]

From #9257 (review):

It seems like having a benchmark that includes this significant cost [of uploading FrameHistory] would be good. Should the benchmark be changed to include zooming?

[stale]

This issue has been automatically detected as stale because it has not had recent activity and will be archived. Thank you for your contributions.