How TiDB benchmarking HTAP hybrid workload?

#1

I tried to run HTAP test in a distributed DB, but I found it hard due to the lack of good hybrid workload. It would be helpful for me to know more about How TiDB Does.

There’re only a few benchmarks for HTAP. CH-benCHmark may be the most popular one, which consists of an entire TPC-C, additional tables and 22 TPC-H equivalent queries.

I found HTAP experientations in TiDB paper (TiDB: A Raft-based HTAP Database, VLDB industry track, 2020, https://www.vldb.org/pvldb/vol13/p3072-huang.pdf). In Chapter6.1, page 9, it goes "Our experiments are conducted under a hybrid OLTP
and OLAP workload using CH-benCHmark. Source code is published online [7]. "

I found source code in CH-benCHmark author’s website https://db.in.tum.de/research/projects/CHbenCHmark/?lang=en, and it made great performance loss when testing (maybe it’s an awful implementation).

So I really want to get further information about TiDB hybrid workload implementation.

  1. How to make EQUIVALENT of 22 queries
  2. How to add additional tables
#2

we used this repo: https://github.com/windtalker/oltpbench
tidb now have tiup bench ch command in tiup tool to run ch-benchmark.

#3

you can also refer to https://github.com/cmu-db/benchbase,
https://github.com/pingcap/go-tpc#ch-benchmark

#4

thanks a lot ! :slight_smile:
I’m going to learn from chbenchmark source code~