Re: [WIP] Zipfian distribution in pgbench

On Wed, Jul 12, 2017 at 12:30 PM, Peter Geoghegan <pg(at)bowt(dot)ie> wrote:
> On Wed, Jul 12, 2017 at 4:28 AM, Alik Khilazhev
> <a(dot)khilazhev(at)postgrespro(dot)ru> wrote:
>> I am attaching results of query that you sent. It shows that there is
>> nothing have changed after executing tests.
>
> But something did change! In the case where performance was good, all
> internal pages on the level above the leaf level have exactly 285
> items, excluding the rightmost page, which unsurprisingly didn't fill.
> However, after increasing client count to get the slow case, the "hot"
> part of the keyspace (the leaf pages owned by the first/leftmost
> internal page on that level) has 353 items rather than 285.

Could you please run the query again for both cases, but this time
change it to get items from the leaf level too, and not just internal
levels? Just remove the "wheretype != 'l' " clause in one of the CTEs.
That should do it.

It's probably the case that the hot part of the B-tree is actually
significantly less than 353 items (or 285 items), and so the "before"
and "after" difference in how many pages are used for the hot part of
the keyspace could be quite a lot larger than my initial estimate. It
could be that the granularity that is shown on the internal pages is
insufficient to see just how bad the problem is.

--
Peter Geoghegan