tcp - RabbitMQ / MQTT performance -


We are using RabbitMQ / with the MQTT adapter as a broker to connect the nodes for monitoring.

Theoretically, if TCP is based, we may have N connections / interfaces such as N port, where N ~ [5000, 65000]. I also realize that due to the many network interfaces, I can theoretically serve more connections.

Is there a sensible limit for the maximum number of connections and hence the number of nodes that are sensible CPU / MEM usage patterns

But, it would also mean to maintain & amp; NN connection processing simultaneously on contact HW (QuadCore 2GHEz / 8GB RAM) Anyone can:

  1. How many connections can be served by the formal process.

  2. What systems / rabbitmq parameters should I look for to optimize?

  3. This will indicate that it is also known by the daemon serving the connection. No formal architecture / design barriers (although at this time we are sticking to RabbitMQ).

At present, 30000 connections can be serviced under config, but% cpu ~ 100% and memory passes around ~ 80 (x300bytes) of messages / seconds Ends with> 60% 

  HW:. I7-3630QM CPU @ 2.40GHz / 4core / 2 thread & amp; 8 GB RAM SW: RabbitMQ ** rabbitmq.config ** {{rabbit, [{TCP_listins, [5672]}, {loopback_user, []}, {vm_memory_high_watermark, 0.8}, {vm_memory_high_watermark_paging_ratio, 0.75} with Ubuntu / MQTT adapter ]}, {Rabbit mmmct, [{default_user, & lt; & Lt; "Guest" & gt; & Gt;} {default_pass & lt; & Lt; "Guest" & gt; & Gt;} {allow_anonymous, true} {vhost & lt; & Lt; "/" & Gt; & Gt;} {exchange & lt; & Lt; "Amq.topic" & gt; {Reciever, true}, {backlog} Apart from this, it is possible to withdraw the retire / connection, based on a policy like cash withdrawal LRU / LFU. But the new connection is for handing over. It can be done on system level (OS / TCP layer). 
 
 Thank you. JB


Comments

Post a Comment