Comments on: A Quick Look into TiDB Performance on a Single Server

By: Attila Nagy

Attila Nagy — Wed, 01 Apr 2020 12:51:17 +0000

What I really miss from these kind of benchmarks are the real world situations. In the real world it’s uncommon to use a database for just inserting/reading some rows. You use a database at least for months, or maybe mor typically years.
It would be nice to have some benchmarks which take this into account: ie they run for months and show how badly these numbers age with that.

By: Roberto Spadim

Roberto Spadim — Wed, 01 May 2019 18:45:27 +0000

shouldn’t be rocksdb vs tidb, instead of innodb vs tidb?

By: Ben

Ben — Fri, 05 Apr 2019 00:40:07 +0000

These benchmarks are interesting but they are kind of Apples to Oranges. That MySQL will win on any write heavy tasks is a sure thing, as both DBs are written with different designs. It will have been nice if you also included a replicated mysql cluster vs TiDB cluster in the results, to see a more real world view. Nobody in there right minds runs a single database instants ( unless you love the risk to lose data)? Another detail left out, … Databases like TiDB, CRDB are more tolerant to run on cheaper / lighter machines thanks to this build in default replication, where as your more or less forced to go with a very reliable ( aka expensive ) server, when faced with a single mysql instance. The lack of a downtime buffer is always present on a single instance mysql server.

By: Michael

Michael — Sat, 02 Feb 2019 02:29:44 +0000

I think the result is misleading.
Some result is even contradict as Øystein Grøvlen pointed out.
What’s more, you don’t show the query execution plan. Maybe there is much room for MySQL to optimize.
TiDB is not suited for OLAP at all, they recommend TiSpark to query complex SQL instead.
As for OLTP, TiDB is no sense.

By: zhao

zhao — Tue, 29 Jan 2019 09:00:02 +0000

I’m interested about the results of PXC & MGR & TiDB on OLTP workload,such as sysbench oltp.

By: Øystein Grøvlen

Øystein Grøvlen — Mon, 28 Jan 2019 02:51:38 +0000

Hi Alexander,
Thanks for sharing this study! What version of MySQL where you using?
I find it a bit strange that MySQL will use more time for the simple count query on the larger machines than on x1e.xlarge. Why should the number of cores matter for single-thread executions? Do you have any explanation for that?
Also, it seems the group by queries uses less time that the simple count on these larger machines. That seems strange to me.

By: Shen Li

Shen Li — Sun, 27 Jan 2019 16:06:47 +0000

Thank you for trying out TiDB! As you note, TiDB is well suited for analytics, but the performance numbers you experienced for the write-only test are lower than expected. This is great feedback, and something we are working on improving in our upcoming 3.0 release.