Re: five-key syscaches
Robert Haas <robertmhaas@gmail.com>
From: Robert Haas <robertmhaas@gmail.com>
To: Yeb Havinga <yebhavinga@gmail.com>
Cc: Joachim Wieland <joe@mcknight.de>, pgsql-hackers@postgresql.org
Date: 2010-07-14T15:04:36Z
Lists: pgsql-hackers
On Wed, Jul 14, 2010 at 10:56 AM, Yeb Havinga <yebhavinga@gmail.com> wrote: > Robert Haas wrote: >> On Wed, Jul 14, 2010 at 7:27 AM, Yeb Havinga <yebhavinga@gmail.com> wrote: >>> Attach is a new patch with all things described above addressed. >> Thanks! >> >> I think we should probably hold off applying this until some of the >> other KNNGIST work is ready, or we have some other concrete need for >> 5-key syscaches. > > Any thoughts about the << 16 and >> 16 bit shifting on the 5th hash key > computation? I blithely copied it from the 3rd key. Hmm, I thought I had the bit in my version, but I see that I don't. Must have gotten lost from an earlier incarnation. It's probably bad to duplicate the bit-shifting pattern of an existing key. We might want to shift by something that's not a multiple of 8, like 12/20. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise Postgres Company