Re: Index Scans become Seq Scans after VACUUM ANALYSE

A Block-sized read will not be rboken up. But if you're reading ina
size bigger than the underlying systems block sizes then it can get
broken up.

So yes a sequential read will get broken up. A single read request for
a block may or may not get broken up. If you're freading with set block
sizes you'll see the set sizes of blocks come through, but what the
underlyign OS does is undefined, same for writing. If the underlying
block size is 8KB and you dump 4MB down on it, the OS may (and in many
cases does) decide to write part of it, do a read ona nearby sector,
then write the rest. This happens when doing long writes that end up
spanning block groups because the inodes must be allocated.

So the write WILL get broken up. Reads are under the same gun. IT all
depends on how big. To the application you may or may not see this
(probably not, unless you're set non-blocking, because the kernel will
just sleep you until your data is ready). Further large read requests
can of course be re-ordered by hardware. Tagged Command Queueing on
SCSI drives and RAIDs. The ICP Vortex cards I use ina number of
systems have 64MB on-board cache. They quite happily, and often
re-order reads and writes when queueing them to keep things moving as
fast as possible (Intel didn't buy them for their cards, they use the
i960 as it is, Intel swiped them for their IP rights). The OS also tags
commands it fires to the ICP, which can be re-ordered on block-sized chunks.

Curt Sampson wrote:

>On Wed, 24 Apr 2002, mlw wrote:
>
>>I am not arguing about whether or not they do it, I am saying it is
>>not always possible. I/O requests do not remain in queue waiting for
>>reordering indefinitely.
>>
>
>It doesn't matter. When they go out to the disk they go out in
>order. On every Unix-based OS I know of, and Novell Netware, if
>you submit a single read request for consecutive blocks, those
>blocks *will* be read sequentially, no matter what the system load.
>
>So to get back to the original arugment:
>
>>>>The supposed advantage of a sequential read over an random read, in
>>>>an active multitasking system, is a myth. If you are executing one
>>>>query and the system is doing only that query, you may be right.
>>>>
>
>No, it's very real, because your sequential read will not be broken up.
>
>If you think it will, let me know which operating systems this
>happens on, and how exactly it happens.
>
>cjs
>