tree f1e732dd594fa85ca55ff7493aaa9decad440bf6
parent 7f3633a59284115e033020492b70ba8ae372b465
author Alex Fandrianto <alexfandrianto@google.com> 1455757756 -0800
committer Alex Fandrianto <alexfandrianto@google.com> 1455757756 -0800

mojo/syncbase: Add Ping Pong Benchmark for Mojo Shell

This benchmark can do normal Ping Pong between syncbases in separate
Mojo shells. The shells may run on Linux or Android. (Combos should work.)
- The syncbase instances mount on the global mount table in "tmp".

To use, see the instructions in the Makefile. (There is a TODO about
improving the usage with a script.) In summary, you can select:
- the number of participating peers
- the sync pattern (Ping Pong, Round Robin, Cycle, PingRandom, etc.)
- the number of iterations to perform for the benchmark
- a benchmark name to avoid test collisions

Here are some example results: (numbers in ms from a full round of PingPong)
- 199.93 for 2-pingpong on Linux (~73 from creator to joiner, and ~128 back)
  Note: Surprisngly, this matches the results from the Go benchmark.

3 devices on Linux
- 257.46 for 3-pingpong on Linux (80, 130, or 180 from creator to joiner, and 120 or 170 back)
- 252.44 for 3-pingall on Linux

Other 3 device patterns (1.5x roundtrips compared to previous patterns)
- 356.56 for 3-cycle on Linux
- 364.94 for 3-reversecycle on Linux
- 383.8 for 3-pingrandom on Linux

And on Android
- ~690 for 2-pingpong on separate Android devices connected to syncslides-alpha

Change-Id: I1dc3d0ac62a3f7453c8974dd1ab078e67af74cc6