From: | Andres Freund <andres(at)anarazel(dot)de> |
---|---|
To: | Robert Haas <robertmhaas(at)gmail(dot)com> |
Cc: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>, Thomas Munro <thomas(dot)munro(at)gmail(dot)com>, pgsql-hackers <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Drop type "smgr"? |
Date: | 2019-02-28 18:11:44 |
Message-ID: | 20190228181144.46a3vq3dbn7tncaq@alap3.anarazel.de |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Hi,
On 2019-02-28 12:36:50 -0500, Robert Haas wrote:
> Well, Andres will probably complain about that. He thinks, IIUC, that
> the buffer tags are too wide already and that it's significantly
> hurting performance on very very common operations - like buffer
> lookups.
Correct. Turns out especially comparing the keys after the hash match is
pretty expensive. It also is a significant factor influencing the size
of the hashtable, which influences how much of it can be in cache.
My plan is still to move to a two tiered system, where we have one
unordered datastructure to map from (db, tablespace, oid) to a secondary
ordered datastructure that then maps from (block number) to an actual
offset. With the first being cached somewhere in RelationData, therefore
not being performance critical. But while I hope to work for that in 13,
I don't think making other large projects depend on it would be smart.
Greetings,
Andres Freund
From | Date | Subject | |
---|---|---|---|
Next Message | David Steele | 2019-02-28 18:13:34 | Re: Remove Deprecated Exclusive Backup Mode |
Previous Message | Jeff Janes | 2019-02-28 18:11:16 | Re: Bloom index cost model seems to be wrong |