> I just took a cursory look at the presentation and Zipkin.io. Would using > Zipkin degrade performance? Would it be considerable? >
In comparison, no. > > Traces (spans) are immediately off-threaded into a Kafka Zipkin transport, and then the Zipkin server has its own Cassandra cluster. This offloads the tracing implementation, through the instrumentation itself always has a small measurable impact. But nothing compared to the 10-20x write amplification of C*'s default tracing implementation.
