I see. If you would ever script benchmark please feel free to share them. It is great to have real user workloads.

Automatic configuration advice would be great and there are actually some work is going on in this area (in the different project however).

Yes there are number of recommended configs available out where. I should probably also post some.

What would be nice is if there could exist some kind of “recommended” values when using mysql administrator.

I mean like a specific page in mysql administrator which based of the output from “SHOW STATUS” etc could hint the admin of which values “might” be good to use for buffers etc from a theoretical point of view (there exist a couple of formulas regarding “recommended” settings in the manual).

I know flupps for example has a couple of “recommended” configs depending of the use of the server in question.

I wish it would be as simple as couple of graphs. In reality so much here depends on workload it would be only missleading to have the graphs. There are also so many variables and some of them are related, so their impact can’t be measure independently so it is truly giant task. I however did some bechmarks with 4.1 looking at performance of DBT2 benchmark under certain workloads. You can find it on our presentations page.

The configs are kind of outdated. It is however something which is just named wrong. The definition of “large” is changing every couple of years. They should be named something like “myisam-mixed-4GB” which does not change over time a lot.

But honestly speaking default values per thread buggers for “large” are already decent. I use same “sample” config file for last 3 years. I typically only modify global buffers (innodb_buffer_pool, key_buffer_size, query_cache_size) to adjust them to size of the box. Most other settings are rather adjusted looking on workload and most typically only give marginal improvement.

Do not forget you also can set values locally, so if you need 64MB sort buffer for one of your queries - you can set it before query is run.

You values I guess are close to what I typically use. I just use smaller read_rnd_buffer_size=8MB and rarely change join_buffer_size as I try to avoid queries which would use it

If you have some benchmarks data that would be great if you could share it.

The 32K case was extreme as well as others. My point was mostly to demonstrate incresing the values do not always increase performance and even can affect it negatively.

Thanks for good comment.

Because I have the impression that the “demo configs” which are included with mysql are somewhat outdated (im thinking of the huge, large etc which are included with the mysql installation for demostration purposes), while my own tests (not that scientific but still :P) shows that a relation of:

sort_buffer_size = 8M
read_buffer_size = 2M
read_rnd_buffer_size = 32M
join_buffer_size = 8M

seems to be a good performance wise start to use the buffers (compared to the default values).

Or is the 32k example in this blog post just a special case of when you perform a count inside a subquery along with a count in the outer query (which is most likely not that common) ?