If the data is compressed, the broker has to recompress the messages in order to assign offsets. So, there is some CPU overhead. However, it shouldn't be too high. How high of the CPU load did you observe?
Thanks, Jun On Mon, Oct 21, 2013 at 12:01 PM, Lu Xuechao <lux...@gmail.com> wrote: > Hi, > > We observed below correlation between kafka configuration and performance > data: > > 1. producer insertion rate drops as compression enabled, especially gzip; > 2. when the producer batch size is below 200, we got high CPU/network IO on > brokers, got high network IO on producers/consumers; > > What's the reasons behind? > > thanks. >