Re: Solaris 9 vxfs equivalent ?
From: Jaime Cardoso (jaimec_at_solsuni.pt)
Date: 11/06/03
- Next message: det: "Re: download probs with x86 u4"
- Previous message: Gary L. Burnore: "Re: Solaris 9 vxfs equivalent ?"
- In reply to: Guy Dallaire: "Re: Solaris 9 vxfs equivalent ?"
- Next in thread: Darren Dunham: "Re: Solaris 9 vxfs equivalent ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Thu, 06 Nov 2003 19:50:23 +0000
Guy Dallaire wrote:
> Well, it looks like I should read the newer documentation. We've always been
> told (oracle doc, white papers from vendors) that RAID-5 was bad for writes.
> Are you telling me that I can put ALL my stuff (except redo) on a big raid-5
> volume and not suffer from write penalty ? There will be updates (sometimes
> massive) to the database and the 3310 only has 512 Mb cache (it's not
> expandable either, alas)
>
No, I don't this that is what you should be doing and, If I understood
correctly, neither is Fredrik.
Raid 5 isn't has bad has it was, It still is a solution cost effective
to give some protection to your Data but, has many things in
performance, it's a trade off
RAID-5 it will always be slower than 0+1/1+0, the thing is that with
caches in the arrays and hardware controlers doing the parity
calculations, thing's aren's has bad has they were.
But, we are getting out of the subject of your initial question, How to
improve HD performance for Oracle?
Vxfs is, usually, a god performance choice for Oracle but, you have to
see for yourself if the extra speed it will bring you to you FS will
compensate for the price and the extra tasks you'll have to do.
You say you have 2* 3310 that will be mirrorer, you don't say how the
array is configured. But, you'll have to make a decision how the Disks
configuration.
One idea is to have a single LUN and place your FS on top of that LUN.
this may be a good idea if you have little concurrent I/O for each FS.
Another idea is to have all 12 Disks available has single disks (OK,
mirrored with the counterpart of the other array) and place, on one the
indexes, on another the tables, redo's on another, etc, etc
this way, each FS will be slower but the concurrency to each Disk will
be smaller.
Of course the solution will be somewhere in the middle but, it's up to
you and your testing to know what is the best setup.
Is you are lucky and your DB isn't in production yet, thet you can do
all the tests needed but, even before that, look into the application
and (if your problem is File system) see How your Application writes and
reads from the disk. Remember the Disks from the 3310 aren't synched (I
think) so, if you are going to write in blocks of 32K to a 2 disk
stripe, you have to see if the best is to use a stripe unit of 16K (and
keeping one disk free to another write request or 32K (and use both
sides of the stripe at the same time.
For short, Vxfs may help you but, it doesn't make your disks to "spindle
faster", look into the file System / LUN design, look into the raid (0,
1, 3, 5, whatever) parameters and, most important of all, look into the
application.
- Next message: det: "Re: download probs with x86 u4"
- Previous message: Gary L. Burnore: "Re: Solaris 9 vxfs equivalent ?"
- In reply to: Guy Dallaire: "Re: Solaris 9 vxfs equivalent ?"
- Next in thread: Darren Dunham: "Re: Solaris 9 vxfs equivalent ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|