Hi,
On 2018-02-20 21:28:40 +0100, Tomas Vondra wrote:
> I don't quite understand why would this case need the TPC-H tests, or
> why would TPC-H give us more than the very focused tests we've already
> done.
Because a more complex query shows the cost of changing cache access
costs better than a trivial query. Simplistic queries will often
e.g. not show cost of additional branch predictor usage, because the
branch history is large enough to fit the simple query. But once you go
to a more complex query, and that's not necessarily the case anymore.
> The first test was testing a fairly short query where any such
> additional overhead would be much more obvious, compared to the TPC-H
> queries that usually do a lot of other expensive stuff.
Unfortunately such reasoning IME doesn't work well with cpu-bound stuff.
Greetings,
Andres Freund