1833559
Members
3381
Online
110061
Solutions
SOLVED
Don,
I was involved with an FC60 implementation about 10 months ago, so i'm a little bit fuzzy about the details but...
As I remember, you are correct in that the FC60 will do RAID0, 0/1 and RAID 5.
I'm not going to comment on performance as this was a new system for us, rather than an upgrade, and its just running SAS rather than Oracle. Also I've never used 12H disk. However, a couple of things to bear in mind:
1. We've had no reliability issues at all with our FC60, or the FC10s we've had on site for the last 12 months.
2. The FC60 fibre channel is generally 5 times faster that FWD SCSI. I seem to remember the FC60 has two FCAL connections to its host (I'm not sure if it can have more than that or whether it now supports Switched fabric rather than just FCAL), and that you see all your LUNs down both channels. Unfortunately LVM doesn't load balance over the two channels so for improved performance remember to alternate which channel is the primary, and which is the standby when adding disks into your volume group. e.g. If you have two LUNS 0 and 1, and you see them as:
/dev/dsk/c2t0d0, /dev/dsk/c3t0d0 LUN 0
/dev/dsk/c2t0d1, /dev/dsk/c3t0d1 LUN 1
Then if you are adding these to a volume group, (assuming they,ve already been pvcreate'd) you would do something like this:
vgextend vg01 /dev/dsk/c2t0d0 (Primary)
vgextend vg01 /dev/dsk/c3t0d0 (Standby)
vgextend vg01 /dev/dsk/c3t0d1 (Primary)
vgextend vg01 /dev/dsk/c2t0d1 (Standby)
This means that both channels will be utilised and you increase your throughput down your 'wire'
3. There's a popular saying: 'Spindles Win Prizes' - that is, it doesn't matter how much on-board cache your disk array has or how fast your IO channels are, the more physical spindles you can get over the most channels into your system, the faster things will go. Look very carefully at the ratio of
GB storage : Spindles : I/O Channels
You might find that rather than using FC60, buying more FC10s and more fibre channel cards provides a better throughput for you. Of course then your dependent on software mirroring/striping so its all relative!
4. Spreading the memory load over several carriers will enhance memory performance, and in fact the N class will tell you during it's boot if memory isn't configured optimally over the memory carriers available. However if disk is your bottleneck this might not actually improve things for you. (particularly if your database is on raw disk, as all database I/O will ignore the buffer cache)
5. I was slightly concerned that you dismissed the archived redo to your FC10, I hopr you still intend to mirror it in the FC10, or write a copy off to tape very quickly after you get it! Writes to the redo log are the only actual writes that Oracle absolutely guarantee will happen, and are therefore very important, after all I can get my datafiles back with redo, I can't get my redo back from my datafiles. I protect my redo at least as well as the rest of the database, if not better.
Hope this helps
/dre
I am an HPE Employee
I was involved with an FC60 implementation about 10 months ago, so i'm a little bit fuzzy about the details but...
As I remember, you are correct in that the FC60 will do RAID0, 0/1 and RAID 5.
I'm not going to comment on performance as this was a new system for us, rather than an upgrade, and its just running SAS rather than Oracle. Also I've never used 12H disk. However, a couple of things to bear in mind:
1. We've had no reliability issues at all with our FC60, or the FC10s we've had on site for the last 12 months.
2. The FC60 fibre channel is generally 5 times faster that FWD SCSI. I seem to remember the FC60 has two FCAL connections to its host (I'm not sure if it can have more than that or whether it now supports Switched fabric rather than just FCAL), and that you see all your LUNs down both channels. Unfortunately LVM doesn't load balance over the two channels so for improved performance remember to alternate which channel is the primary, and which is the standby when adding disks into your volume group. e.g. If you have two LUNS 0 and 1, and you see them as:
/dev/dsk/c2t0d0, /dev/dsk/c3t0d0 LUN 0
/dev/dsk/c2t0d1, /dev/dsk/c3t0d1 LUN 1
Then if you are adding these to a volume group, (assuming they,ve already been pvcreate'd) you would do something like this:
vgextend vg01 /dev/dsk/c2t0d0 (Primary)
vgextend vg01 /dev/dsk/c3t0d0 (Standby)
vgextend vg01 /dev/dsk/c3t0d1 (Primary)
vgextend vg01 /dev/dsk/c2t0d1 (Standby)
This means that both channels will be utilised and you increase your throughput down your 'wire'
3. There's a popular saying: 'Spindles Win Prizes' - that is, it doesn't matter how much on-board cache your disk array has or how fast your IO channels are, the more physical spindles you can get over the most channels into your system, the faster things will go. Look very carefully at the ratio of
GB storage : Spindles : I/O Channels
You might find that rather than using FC60, buying more FC10s and more fibre channel cards provides a better throughput for you. Of course then your dependent on software mirroring/striping so its all relative!
4. Spreading the memory load over several carriers will enhance memory performance, and in fact the N class will tell you during it's boot if memory isn't configured optimally over the memory carriers available. However if disk is your bottleneck this might not actually improve things for you. (particularly if your database is on raw disk, as all database I/O will ignore the buffer cache)
5. I was slightly concerned that you dismissed the archived redo to your FC10, I hopr you still intend to mirror it in the FC10, or write a copy off to tape very quickly after you get it! Writes to the redo log are the only actual writes that Oracle absolutely guarantee will happen, and are therefore very important, after all I can get my datafiles back with redo, I can't get my redo back from my datafiles. I protect my redo at least as well as the rest of the database, if not better.
Hope this helps
/dre
I am an HPE Employee
