Hi Chen Chen,
Thanks for checking! No, I have not applied any custom configurations apart from the number of threads in the tools. For example, with recent myloader version you can also disable the redo log which can bring a huge improvement as well. In case you want more information, you can take a look at my presentation that I made about this topic:
https://www.slideshare.net/ViniciusMalvestioGri/pl22-backup-and-restore-performancepptx

Thanks for your sharing.

Did you disable binlog or enable binlog for util.loadDump and myloader when load data.

the default value of related option is not the same.

for util.loadDump, skipBinlog defaults to false.

but for myloader, if we don’t set –enable-binlog, binlog is turned off by default.

Hi George,

No I have not tested this option. Even it offers some potential for parallelization, I do not believe it is commonly used. And given that there are better tools to achieve parallelization (even in MySQL community binaries), I would not recommend it.

But yes, in such way, tweaking these settings should bring some improvement when compared to the default settings of mysqldump.

Hi Kenny,

I believe the best would be to write another post dedicated to maximize data ingestion into MySQL. Posting all these metrics here could make the post bloated.

In either case, it will be a nice comparison from MySQL shell -> MySQL shell and myloader -> myloader when using such features.

Thanks for your response Vinicius. Nice article bdw.

Disabling redo log and binary log will improve performance a lot, not just a couple of % difference, but potentially a few times faster. In some tests we have done internally we have observed 1GB/s loading performance (of course that is schema & data dependent & beefy hardware though!).

It is my opinion that when performance is compared, it’s important to use these simple settings that drastically improve performance. It doesn’t need to be fully tuned, but the low hanging fruit should at least be covered.

I’m not trying to convince mydumper is slower than MySQL Shell Dump/Load, I don’t really know in this test. The parallel tools mysql Shell Dump/Load & mydumper will be a lot faster, and loading a mysqldump/pump will likely not.

I know a lot of users read Percona’s Blog and many of them would benefit from knowing these major performance increase that can be obtained for loading logical data. (And it would also be nice to see what performance the Percona experts observe)

Hi gryp,

Both were enabled at the time of the test. In my analysis, I mention that it is possible to extract more performance for the load phase (including a few links). What I tried to do is to cover the regular scenario, not the best case scenario.

Probably using such settings, MySQL shell utilities or mydumper/myloader would be the fastest tool.

My idea is not elect a champion in this test, but provide the whole range of options that DBAs can have and its potential for optimization. For sure, even mysqldump would run faster but the main issue is its single thread nature.

Was the redo log and binary log enabled in this test?
Disabling both of these features will make any logical loading much faster.

Hi Ankit,

When we think about creating a new replica, both ways can achieve the result (physical and logical backups). When we consider the process (backup/restore) it makes sense to compare. Also, because of the popularity of the tool, it is important to give an idea of how fast the tool can be.