This page contains results comparing signal send and receive throughput before and after the parallel signal send optimization. See the commit message of this commit for more information about the optimization.
The benchmark lets N Erlang processes send signals (of configurable types and sizes) to a single process during a period of T seconds. Both N and T are configurable variables. A signal with size S has a payload consisting of a list of length S with word-sized (64 bits) items. The send throughput is calculated by dividing the number of signals that are sent by T. The receive throughput is calculated by waiting until all sent signals have been received, and then dividing the total number of signals sent by the time between when the first signal was sent and when the last signal was received.
Machine:
Microsoft Azure VM instance: Standard D64s v3 (64 vcpus, 256 GB memory):
Operating System:
Description: Ubuntu 18.04.2 LTS Linux version: 5.4.0-1051-azure
The benchmark was started with the parameter "+sbt tnnps".
erl +sbt tnnps -eval "parallel_messages_SUITE:large_throughput_benchmark(),erlang:halt()"
The benchmark configuration used can be found here.
Notice that every second graph contain send throughput and every second contain receive throughput.
This page generates graphs from data produced by the Message Send/Receive Benchmark which is defined in the function parallel_messages_SUITE:test_throughput_benchmark/1
(see "$ERL_TOP/erts/test/parallel_messages_SUITE.erl
").
Note that one can paste results from several benchmark runs into the field below. Results from the same scenario but from different benchmark runs will be relabeled and plotted in the same graph automatically.
Note also that that lines can be hidden by clicking on the corresponding label.
Paste the generated data in the field below and press the Render button: