File Striping for Oracle Database

[Logan Ramasamy_1]

I am a DBA. I am not a HP guru. I was listening to a presentation by Tom Kyte, VP, Oracle in the internet. He said, "Prior to RAID arrays and file system striping, DBAs attempt to spread I/O out by spreading data and index files. But now, underlying file system is striped; One big mount point is carved out of hundreds of physical devices, underneath. I/O is spread across all of those devices evenly.". I asked our system Administrator to comment on this with respect to our environment. Her answer was:"This is correct, however on the HPUX systems we do not use RAID, we use mirrors.". I did not understand this. As far as I understand, mirroring is to duplicate one physical disk by another. If a disk fails, repalce it. It will sync itself with the O.K. one. How this mirroring meets the requirement of file system striping? Please help me to understand this in a high level plain English. Thanks.

[Sandman!]

What kind of storage is your data on...JBOD or high-end arrays like XP's?

[Christian Tremblay]

If your storage sits on a RAID disk array, then volume disk striping makes no sense since the array itself will manage where on the physical disks your data will be written.

She may have been telling you that you don't have intelligent disk arrays in your environment then striping over multiple disks makes sense. Mirroring is just writing to 2 or more physical disks at once.

[Logan Ramasamy_1]

I can go back to her and find out what kind of storage is ours. But I doublt if I get any response which will make sense to me. It is a commin disk array shared between 2 HP boxes (HP-UX B.11.11 U 9000/800).

[Alzhy]

Logan,

INSIST that you as a DBA knows the storage infrastructure -- i.e. your Storage Arrays and how the "LUNS/Disks" are RAIDed. Also, whan a storage/system admin refer to a LUN/disk as mirrored -- it's just another level of RAIDing -- but does that not neccessarily mean you have striped LUNs.

Depending on your Storage Array -- it is always best to know its architecture as sometimes teh BEST I/O performance and scalability can be achieved by HOST BASED STRIPING (via LVM/VxVM, etc) of these already RAIDed LUNs/disks from yout array.

[Hein van den Heuvel]

I suspect a simple communication problem.

You are right, the goal of mirrorring is just that, mirror the data to keep it available in case a drive fails. It also tends to have performance benefits as reads (which tend to dominate the IO pattern) can come from the nearest, or least busy, disk. Writes may slow down, as both disks need to be written, but caches can take care of that.

The system administrator may have focussed on the mirrored aspect of a lun, while it offers striping at the same time. This combination is often refererred to as raid-01 or raid-10. It offers the best of both worlds.

Ask you question in simple terms... how many physical disks are behind my luns?
2 would be mirroring only and not good enough. 8 would be nice. 40+ would be the 'SAME' architecture Tom referred to.

>> I can go back to her and find out what kind of storage is ours. But I doublt if I get any response which will make sense to me.

How can that be? The SA is there to serve the DBA, which is there to serve the application no? If the SA can not communicate with the DBA then you can not run an optimal business solution.

Hope this helps some,
Hein van den Heuvel (at gmail dot com)
HvdH Performance Consulting

[Yogeeraj_1]

hi Logan,

RAID, striping, mirroring are not just about HP systems - they are general concepts.

The following URL answers many of the questions people are asking about RAID for an Oracle database:
http://technet.oracle.com/deploy/availability/pdf/oow2000_sane.pdf

hope this helps!
kind regards
yogeeraj

[Steven E. Protter]

Shalom,

It is much better to handle this on a disk array and merely present a well configured LUN to the HP system.

The disk array is designed for this, the system is not as good at making the i/o most efficient.

Striping such that it is with the OS is not very effective and not even considered true striping.

SEP

[Ralph Grothe · Accepted solution]

You can try to find out for yourself
if you don't feel inclined to ask your Unix sysadmin who should know.
First add /usr/sbin to your path to abbreviate typing.
(assuming your dba account has a Bourne-like shell)

$ PATH=/usr/sbin:$PATH


With sqlplus you should be able to find out $ with an SQL query the paths to the *dbf files (if you don't know already, anyway).
Sorry, I would have to look up the correct SQL statement as I am no DBA.

Then bdf should show you the logical volume (LV) of the mount that bears the filesystem.

$ bdf /path/to/some/file.dbf

The LV path of the bdf will contain the volume group (VG) that LV belongs to.

To find out what disks are contained in that VG you can run verbose vgdisplay.

$ vgdisplay -v vgXX | more

At the end will be listed the physical volumes (PV) aka disks.

$ vgdisplay -v vgDat1|awk '/PV Name/&&!/lternate/{print$NF}'

You also can pipe this to lssf which should disclose the HW paths of the disks

$ vgdisplay -v vgDat1|awk '/PV Name/&&!/lternate/{print$NF}'|xargs lssf

By the HW paths you are likely to discover how the used disks are attached.

Compare the beginning of HW paths to output of

$ ioscan -knfCext_bus

to find out the controllers the disks are attached to.

You may also follow a whole HW path by

e.g. (replace with your HW path)

$ ioscan -knf -H 1/0/6/1/0.114.4.19.0

To find out if any of the LVs are striped you can do this.

$ $ vgdisplay -v vgXX|awk '/LV Name/{print$NF}'|xargs lvdisplay|grep -E 'LV Name|Stripes'


HTH

[Logan Ramasamy_1]

Thanks everybody who took sometime to part with knowledge. I have attached the output for the unix commands. I would appreciate very much if you help me to conclude my original question. ( I know you guys are real Unix gurus. Some of the names are familar to me).