Re: Software RAID and drive failures
On Tue, 2003-03-11 at 14:30, Kirk Strauser wrote:
> At 2003-03-11T18:41:12Z, Ron Johnson <ron.l.johnson@cox.net> writes:
>
> > Why wouldn't you get parallel reads, especially if you do async IO?
>
> I guess a better example would be when two files are accessed, and each
> has blocks that physically reside on the same two drives. Imagine that both
> files are accessed randomly and concurrently (think multiple database
> instances). In this case, it could very well happen that blocks from file1
> (f1a, f1b) and file2 (f2a, f2b) are being read simultaneously. In this
> case, both drives are jumping back and forth between f1a/f2a and f1b/f2b.
2 issues:
1. Tagged command queuing should order the reads.
2. As a DBA, it seems to me that this possibility is irrelevant, since
the DBMS would have sucked those blocks (and their neighbors) into
it's cache after the 1st 1 or 2 db read requests.
> Imagine, instead, that file1 and file2 are on a mirror. The RAID system
> could converge on the situation where the first drive is "dedicated" to
> file1, and the second is "dedicated" to file2.
>
> I don't think this is a particularly unlikely scenario.
Are controllers smart in that way?
--
+---------------------------------------------------------------+
| Ron Johnson, Jr. mailto:ron.l.johnson@cox.net |
| Jefferson, LA USA http://members.cox.net/ron.l.johnson |
| |
| Spit in one hand, and wish for peace in the other. |
| Guess which is more effective... |
+---------------------------------------------------------------+
Reply to: