[rabbitmq-discuss] RabbitMQ QUEUE MESSAGES
NuZajacPogodi at gmail.com
Fri Jul 12 21:37:07 BST 2013
Ok, so it should be taken the most pessimistic condition - the queue fully
paged on disk - the performance measured and now it can be stated that
Rabbit has O(1) queue processing complexity but with a very low message
processing rate. Even having a lot of RAM messages processing rate drops
drasticly after a small fraction of RAM is occupied by queues since paging
causes Rabbit to be disk io bound.
That is great that you are treating it as feature to provide more consistent
queue throughput by optionaly preventing paging. The system will be much
easier to analyse and plan for a deployment. Having a given message size and
using limited-length queues it will be possible to utilize RAM resources to
higher extent with a constant and high message processing rate.
It is espetially usefull in scenarios when message producer is driven by
events (sometimes storm of events) from external world and Rabbit is used as
a buffor for storing tasks for later processing by other less efficient
consumers. In that situation it is undesirable Rabbit to slow down producers
- producers would need to implement their own buffer because they cannot
stop incoming events - an addtional buffer to protect the other buffer does
not look elegant.
View this message in context: http://rabbitmq.1065348.n5.nabble.com/RabbitMQ-QUEUE-MESSAGES-tp27737p28065.html
Sent from the RabbitMQ mailing list archive at Nabble.com.
More information about the rabbitmq-discuss