FC60 - upgrading all disks

[Marilyn DuChene]

We have an FC60 w/ 2 SC10 enclosures on a N class running HP-UX 11.0. We will be replacing all the disks with larger ones (18 to 36gb). We will be using RAID 0/1 and we will have 2 luns and a split bus config. I have 2 questions -
1. What segment size should we set? I see in the manual it suggests 64K - we currently have 8K seg size & a cache page size of 4K. Should I increase the segment size?
2. This is the steps I plan to execute to do the upgrade - is anything missing?
1. backups
2. umount, vgchange, vgexport vg's affected
3. unbind the luns
4. power down contoller & then sc10s
5. replace all disks
6. power up sc10's & then the controller
7. bind luns
8. insf -e
9. rebuild file systems

Thanks to all,
Marilyn

[harry d brown jr]

Marilyn,

I'd increase the segment size, but it really has to do with your data layout. If 64K is optimum, then bump it up.

Also, you need a step 10: Restore data :-))

live free or die
harry

[Vincent Farrugia]

Hello,

The segment size optimisation depends on the data you have. This is for performance issues more than anything else.

Your hard disk upgrade method and steps are good. You could also replace harddisks one at a time allowing for rebuild, but that would mean a week's time of rebuilding :-)

HTH,
Vince

[Sandip Ghosh]

The data size is totally depending on your type of usage. If you are having mostly datafiles in your Array , then you can make it 64K. But if the no of files are higer proportional to size of the file then I would go for 8K size.

Sandip.

[Sandip Ghosh]

Actually HP is recomending this to allow you to put maximumdisk space in a volume group. The Max pe per PV is maximum 65535. So if you keep the pe size as 64K you can allocate more space in a VG rather than if you keep the pe size as 4K.

Sandip

[Marilyn DuChene]

I wasn't considering changing the PE Size of my vg's. I didn't think there was a correlation between segment size of the disk array to the PE size of you vg. Maybe they should be the same? We have Universe database with maybe 25% very large files, and the rest small. Maybe a mixed bag of stuff. If anybody out there has any other input - would like to hear it - otherwise, we might go with our current set up (8k seg size) since it didn't cause us any grief.

[Insu Kim]

In the manual,
Optimum performance is typically achieved when the segment size
matches the I/O size. In this case, only one disk is required to service an I/O, leaving the
remaining disk in the LUN available for other I/Os.
A large segment size provides good read performance in most RAID levels. The controller will
have to access fewer disks to retrieve the data, leaving the remaining drives available for other
I/O operations. Large segment sizes are typically useful for applications that require high I/O
throughput.
A small logical unit segment size is useful for most RAID 5 write applications, because the
controller firmware is capable of performing group writes (writing of data simultaneously to
multiple disks, while calculating the parity for the stripe, as opposed to the single-threaded read-modify-
write). Small segment sizes are typically useful for applications that require large
numbers of small I/Os to be processed quickly.