The most important thing is to have enough I/Os for writes. If you configure a broker to do 1 flush to disk every n messages and each flush takes x ms. Then you can produce about n * 1000 / x messages per second. This is for one topic/partition. If there are more topics/partitions, there will be more parallel I/Os and the supported throughput will be larger.
Jun On Thu, Jan 5, 2012 at 7:20 AM, Irfan Mohammed <irfan...@gmail.com> wrote: > Hi, > > Kafka docs seem to indicate that the system has low CPU and memory > requirements. What is an ideal configuration and recommendations for the > system? > > Any straw man formulae based on the number of messages per second, expected > message latency and delivery modes? > > Any working production specs would greatly help. > > Thanks, > Irfan >
