Sequential Scan Read-Ahead

At 12:41 PM 4/23/02 -0400, Bruce Momjian wrote:

>This is an interesting point, that an index scan may fit in the cache
>while a sequential scan may not.

If so, I would expect that the number of pages read is significantly
smaller than it was with a sequential scan. If that's the case,
doesn't that mean that the optimizer made the wrong choice anyway?

BTW, I just did a quick walk down this chain of code to see what happens
during a sequential scan:

access/heap/heapam.c
storage/buffer/bufmgr.c
storage/smgr/smgr.c
storage/smgr/md.c

and it looks to me like individual reads are being done in BLKSIZE
chunks, whether we're scanning or not.

During a sequential scan, I've heard that it's more efficient to
read in multiples of your blocksize, say, 64K chunks rather than
8K chunks, for each read operation you pass to the OS. Does anybody
have any experience to know if this is indeed the case? Has anybody
ever added this to postgresql and benchmarked it?

Certainly if there's a transaction based limit on disk I/O, as well
as a throughput limit, it would be better to read in larger chunks.

cjs
--
Curt Sampson <cjs(at)cynic(dot)net> +81 90 7737 2974 http://www.netbsd.org
Don't you know, in this new Dark Age, we're all light. --XTC