diff options
| author | Matthew Hodgson <matthew@matrix.org> | 2017-08-15 18:27:42 +0100 |
|---|---|---|
| committer | Matthew Hodgson <matthew@matrix.org> | 2017-08-15 18:27:42 +0100 |
| commit | 92168cbbc53ccf941ddcb958452ace8e41a948fd (patch) | |
| tree | 35f58745d0d4a99adbcf4208aa95ec9129d2d7b2 /synapse/config/server.py | |
| parent | Merge pull request #2415 from matrix-org/rav/synctl_cpu_affinity (diff) | |
| download | synapse-92168cbbc53ccf941ddcb958452ace8e41a948fd.tar.xz | |
explain why CPU affinity is a good idea
Diffstat (limited to '')
| -rw-r--r-- | synapse/config/server.py | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/synapse/config/server.py b/synapse/config/server.py index 4e4bf6b432..e33cd51f7c 100644 --- a/synapse/config/server.py +++ b/synapse/config/server.py @@ -153,10 +153,18 @@ class ServerConfig(Config): # bit corresponding to the first logical CPU and the highest order bit # corresponding to the last logical CPU. Not all CPUs may exist on a # given system but a mask may specify more CPUs than are present. + # # For example: # 0x00000001 is processor #0, # 0x00000003 is processors #0 and #1, # 0xFFFFFFFF is all processors (#0 through #31). + # + # This is desirable for Synapse processes (especially workers), which are + # inherently single-threaded due to the GIL and can suffer a 30-40% slowdown + # due to cache blow-out and thread context switching if the scheduler happens + # to schedule the underlying threads across different cores. + # See https://www.mirantis.com/blog/improve-performance-python-programs-restricting-single-cpu/ + # # cpu_affinity: 0xFFFFFFFF # Whether to serve a web client from the HTTP/HTTPS root resource. |