Re: How many of you use S.A.M.E?
- From: "Mark Brinsmead" <pythianbrinsmead@xxxxxxxxx>
- To: ax.mount@xxxxxxxxx
- Date: Fri, 2 Feb 2007 21:13:29 -0700
Amonte,
It is, in general, possible to decide at least approximately where data
will go on the disk. The most obvious method to do this is with
partitioning. When you create a disk partition (either to support a
filesystem or provision a raw partition) you specify the range of
blocks/tracks that will be used. This is a common sysadmin task, but many
DBAs never have do this themselves. Perhaps that is why you are unaware of
the technique.
No sane person has ever (to my knowledge) tried to place specific data on
specific blocks, but many people can and do arrange to put certain
classesof data in designated
regions of a disk. That doesn't mean nobody has ever asked me to put
specific data on specific block, just that those who did probably weren't
sane. But I digress.
A very long time ago, and with certain operating systems, you could
contrive to do this sort of thing. Used to be, some DBAs would even do
things like try to place index blocks on one specific disk surface, and the
"corresponding" data blocks on the physically opposite disk surface (and
track and block). But that was quite some time ago -- in many cases, you
can't even determine how many surfaces a device actually has, and the
question of how many blocks there are in each track is almost completely
meaningless... (Unless you are prepared to take into account that the
answer is usually different for each track, anyway.)
That said, in many cases, you can still specify things like "online redo
goes on the outer 5%; indexes on the next 20%; ...". More or less, anyway.
In reality, though, today's storage devices are getting more and more
"intelligent" (or sometimes deceitful), that is making this harder and
harder to do. Sometimes, at least...
Personally, I do not have any experience with EMC "Meta Devices", but
based on the description I could definitely imagine them holding some
unpleasant surprises for people laying out databases unwarily... This can
include errors like mistaking "meta devices" for physical disks and doing
RAID-0 striping across them, or even naively multiplexing online redologs
across metadevices that happen to be stored on the same physical devices.
While you don't need to know precisely where a given block is located on
a disk, if you do want to plan your I/O you do need to know how your
database files map onto physical devices...
On 2/1/07, amonte <ax.mount@xxxxxxxxx> wrote:
Hi Brandon
Have you used EMC Meta Devices before?
By the way have you tried by chance what Loaiza suggested, putting data in
specfic physical sectors of a hard drive? I am really curious how can that
be achieved. S.A.M.E is simple but putting data as he says makes life
impossible dont you think so?
Regarding query tuning, it is good, perfect. However the real world
experience has taught me that, if you have 50 lousy or even experienced
old developers and 4 DBAs you can spend 100% of DBA time tuning queries and
it will still be endless. My opinion is we have to get some sort of balance,
not be too biased to any.
--
Cheers,
-- Mark Brinsmead
Senior DBA,
The Pythian Group
http://www.pythian.com/blogs
Other related posts:
