A few years back Deva wrote about how to use tcpdump on very busy hosts. That post sparked my interest about exploring how to measure the impact of tcpdump on very busy hosts. In this post, I wanted to highlight how much of an impact there really is and what options you have to make the query collection much more effective.
Some things you need to know:
- The test is a sysbench read-only workload, 8 tables, 8 threads, 1000000 rows each with 16G of buffer pool. Dataset fully in memory.
- sysbench is ran on the same host, on 1Gbps connection, sysbench can saturate the network and therefore affect my network test with netcat so I decided to run locally.
- There are 13 tests, 5 minutes each with 1 minute interval,
varying on how the dump file is captured.
- First one as …