Hello, I'm trying to optimize RabbitMQ instance for our environment (Large number of connections, few messages).
Thus, I tested various tcp_listen_options.sndbuf & recbuf.
For instance, I set snd & rec buffer size very low for the test.
- mqtt.tcp_listen_options.sndbuf = 1
- mqtt.tcp_listen_options.recbuf = 1
Also, I set the & rec buffer size very high for the test.
- mqtt.tcp_listen_options.sndbuf = 212992000 (~ 200MB)
- mqtt.tcp_listen_options.recbuf = 212992000
For both setups, I connect 100 concurrent MQTT clients & check the memory consumed.
However, consumed total memory (calculated by RSS strategy) was almost the same for both options!
I notice that the memory consumption was different in both cases when the calculation strategy was 'allocated' (1MB vs 20GB memory consumption for the above cases).
However, that does not affect the system performance since it was only virtually allocated memory not physically used by the system (I checked with 'top' command).
In conclusion, I can't see the effect of resizing TCP buffer size for reducing memory consumption.
Are there any configurable options that I forgot to use? Then, please let me know.