Video Buffer Performance - Low Level Spec

Overview

The goal of this function is to get the instances for a given frame list.

In general for this function we expect

  1. Nearly 100% index hit on DB side
  2. To use the Diffgram File Instance Cache concept.
    The cache is assumed to be valid on read. Systems that write to that are assumed to be responsible for setting cache dirty.

Expected Performance

Times listed are from client view "end to end" times.

For a batch of 60 frames with 8 instances per frame it should return in less then 500 MS.

For a batch of 60 frames with 80 instances in each (4,800 instances) we expect the end to end function to return in less then 2000 MS in most cases.

28,800 instances (360 frames) loads in about 8000 MS.

These numbers are meant as a rough baseline to gauge if this subsystem is behaving as expected.

Better Performance

For improved performance, note that you can make multiple parallel calls, using different frame ranges and keys.

If this is a bottleneck for your use case please open an issue on github.

Note that the front end may pre-load instances, so this time may not be directly visible to the end user.


Did this page help you?