Autobahn WebSocket Echo server on multicore
This example demonstrates how to scale-up an AutobahnPython based WebSocket echo server on a multicore machine.
Usage
Run the server with 4 workers:
pypy server.py --wsuri ws://localhost:9000 --workers 4
In general, using more workers than CPU cores avaiable will not improve performance, likely to the contrary.
Detailed usage:
$ pypy server.py --help
usage: server.py [-h] [--wsuri WSURI] [--port PORT] [--workers WORKERS]
[--noaffinity] [--backlog BACKLOG] [--silence] [--debug]
[--interval INTERVAL] [--profile] [--fd FD] [--cpuid CPUID]
Autobahn WebSocket Echo Multicore Server
optional arguments:
-h, --help show this help message and exit
--wsuri WSURI The WebSocket URI the server is listening on, e.g.
ws://localhost:9000.
--port PORT Port to listen on for embedded Web server. Set to 0 to
disable.
--workers WORKERS Number of workers to spawn - should fit the number of
(phyisical) CPU cores.
--noaffinity Do not set worker/CPU affinity.
--backlog BACKLOG TCP accept queue depth. You must tune your OS also as
this is just advisory!
--silence Silence log output.
--debug Enable WebSocket debug output.
--interval INTERVAL Worker stats update interval.
--profile Enable profiling.
--fd FD If given, this is a worker which will use provided FD
and all other options are ignored.
--cpuid CPUID If given, this is a worker which will use provided CPU
core to set its affinity.
Load Testing
You will need some serious WebSocket load driver to get this thingy sweating. I recommend wsperf for various reasons. wsperf is a high-performance, C++/ASIO, multi-threaded based load driver. Caveat: currently, even when using wsperf, the bottleneck can still be wsperf when running against Autobahn/PyPy. You should give wsperf more CPU cores than Autobahn for this reason.
You should also test on non-virtualized, real-hardware and you will also need to do OS / system level tuning, please see the instructions here.
For results, please see here.