Re: Clock sweep not caching enough B-Tree leaf pages?
Peter Geoghegan <pg@heroku.com>
From: Peter Geoghegan <pg@heroku.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: Jim Nasby <jim@nasby.net>, Robert Haas <robertmhaas@gmail.com>,
Andres Freund <andres@2ndquadrant.com>, Merlin Moncure <mmoncure@gmail.com>,
Pg Hackers <pgsql-hackers@postgresql.org>
Date: 2014-04-22T00:42:14Z
Lists: pgsql-hackers
On Mon, Apr 21, 2014 at 5:28 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > We used to have one. It was a big bottleneck --- and this was years > ago, when the buffer manager was much less scalable than it is today. > (IIRC, getting rid of a central lock was one of the main advantages > of the current clock sweep code over its predecessor.) Yes, it was. This is a major advantage of clock sweep, and anything that replaces it will need to maintain the same advantage. Didn't someone indicate that clock sweep could beat ARC around that time, presumably for this reason? If no one did, then my reading of a variety of other papers on caching indicates that this is probably the case. -- Peter Geoghegan
Commits
-
Replace the BufMgrLock with separate locks on the lookup hashtable and
- 5d5087363d7c 8.1.0 cited