maria-discuss team mailing list archive

[Daniel Black <daniel@xxxxxxxxxxx>]

On Fri, Dec 16, 2022 at 12:44 PM Jonathan M. Wilbur <jonathan@wilbur.space>
wrote:

> Sorry to ask this question here, since it is not strictly MariaDB-related,
> but I need the knowledge of people that develop with databases: is there
> significant overhead to using a DBMS versus a storage engine directly?
>

DBMS add a lot of query functionality, enforces ACID constraints, but also
tries to make generalized tradeoffs favouring read performance and
consistency and doing its best to minimize the write speed impacts.

So in general you're better picking a solution than rolling your own.

>
> I have a project for which speed / latency is very important,
>

Read speed or write speed? What about durability?

Is the read access on recent writes or archival? Are reads complex?


> and I was wondering if I could make significant performance gains by using
> a storage engine (such as WiredTiger, RocksDB, LevelDB, etc.) directly
> instead of using a DBMS (like MariaDB). More concretely, are there any
> studies out there on this to give me some objective numbers on this?
>

The MariaDB layer aims to be as minimal as possible.

The only test case that really matters is your workload. I'd start modeling
the client as a benchmark on this and see how it goes on MariaDB. See if
the numbers seem fair. Modelling an alternate implementation should be easy
after this.


> Assume it is over a local loopback connection, so pretty much only the
> overhead of the networking itself would apply.
>
> Thank you in advance,
>
> - Jonathan M. Wilbur
> _______________________________________________
> Mailing list: https://launchpad.net/~maria-discuss
> Post to     : maria-discuss@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~maria-discuss
> More help   : https://help.launchpad.net/ListHelp
>