← Back to team overview

drizzle-discuss team mailing list archive

Re: Drizzle Query processing modules shortterm/longterm roadmap ?


Prafulla Tekawade <prafulla_t@xxxxxxxxxxxxxxxxxxxxx> writes:

> They had "vectorised" query operators (All operator would execute
> on a batch of tuples rather than a single tuple.) It gave really lot
> of good performance improvement because it made better use of
> CPU cache.
> We can have something like that in drizzle.

This makes me think of BKA, "batch key access" in MySQL. This is a feature
where the optimiser can send batches of key values into the storage engine, and
the storage engine will return batches of matching tuples. This is used eg. in
nested-loop joins to work on batches of tuples at a time.

Not sure if this is what you had in mind though.

I'm actually also not sure which tree this currently exists in, if any, though
I know it exists (I re-wrote the NDB handler to use the new BKA interface at
some point). Could be that it was in MySQL 6.0 (and so perhaps even in
Drizzle), but I vaguely recall that it was never in main MySQL 6.0, only a 6.0
feature tree. It might be in MariaDB 5.3. A question to the optimiser guys in
#maria on FreeNode will probably provide a clear answer in case of interest.

 - Kristian.

Follow ups