I've been testing the performance between 5.7 and 8.0. We figured out that our application is around 30% slower with 8.0, so I tried to narrow down the issue. I've started running simple sysbench tests and they showed the same perf hit. Then I found this nice article.
I tried to replicate the setup (4 cores, NVMe backend, only VM on the host, etc) the best I can. I used two separate VMs based on CentOS 7 with community editions. However, after tests I got the same ~30% perf hit with the same scripts found in the blog post. Then I tried to add the my.cnf options mentioned in the blog, nothing. Apart from these options, my.cnf was left as default (except innodb_buffer_pool, which was set to 8G from the start and binlogs were disabled on both). As a last resort I just tried a simple sysbench run (removed all the extra options, ran it for 10 minutes), same thing. I even tried to force M8 to use latin1 instead of utf8 (while creating the DB), which yielded these (best) results: Results
Everything points to M8 having way less performance as M5.7, which contradicts ALL google listing.
Does anyone have any idea where the perf hit comes from?
Thanks in advance!


