Solaris

Hi,

I am on Solaris 9. I have 4 disks that I want to mount in RAID 1+0 with LVM
First I initialize the devices ; for example :

metainit d102 c1t2d0s6
metainit d103 c1t3d0s6
metainit d104 c1t4d0s6
metainit d105 c1t5d0s6

then I make the mirrors :

metainit d120 -m d102
metattach d120 d104
metainit d121 -m d103
metattach d121 d105

and least, I want to make the stripe with :

metainit d130 1 2 d120 d121 -i 64k

and the system answers : d120 is a metadevice. metainit seems to want only a physical component !

Is it impossible to implement this solution with LVM ?

Thanks if somebody could help me.

See Sun Docs - DiskSuite user guide creating a stripe metadevice. After you create the stripe sets, then you mirror.

Yes.
I see what you mean. But by this way I do RAID 0 + 1, not RAID 1 + 0 which give a better security.

To create a RAID 0+1 metadevice, one creates two striped metadevices and then one creates a mirror of these metadevices; this is standard practice with DiskSuite.
Creating a RAID 1+0 device requires creating multiple mirrors of disk partition pairs, then striping together the mirror metadevices. The problem is that DiskSuite does not allow the striping of metadevices. Members of a striped metadevice must be physical disk slices, hence RAID 1+0 appears not to be possible with DiskSuite.
However, things are not always as they appear.

If the RAID 0+1 metadevice is created such that the number of columns in each of the mirrored stripes is the same, and the sizes of the physical disk slices in each column is the same, DiskSuite will internally convert that metadevice to a RAID 1+0 device. However, no difference in its outward appearance (for example, using the metastat command) will occur.

As an example, give you what looks to be a RAID 0+1 metadevice:

Code:

  d0: Mirror
       Submirror 0: d10
         State: Okay         
       Submirror 1: d20
         State: Okay         
       Pass: 1
       Read option: roundrobin (default)
       Write option: parallel (default)
       Size: 7058880 blocks

   d10: Submirror of d0
       State: Okay         
       Size: 7058880 blocks
       Stripe 0: (interlace: 32 blocks)
           Device     Start Block  Dbase        State Reloc Hot Spare
           c1t0d0s0          0     No            Okay   Yes 
           c1t1d0s0          0     No            Okay   Yes 
           c1t2d0s0          0     No            Okay   Yes 

   d20: Submirror of d0
       State: Okay         
       Size: 7058880 blocks
       Stripe 0: (interlace: 32 blocks)
           Device     Start Block  Dbase        State Reloc Hot Spare
           c2t0d0s0          0     No            Okay   Yes 
           c2t1d0s0          0     No            Okay   Yes 
           c2t2d0s0          0     No            Okay   Yes

However, since DiskSuite treats this device as a RAID 1+0, non-corresponding disks in each submirror CAN fail without any resulting data loss. To illustrate this point, if the disks c1t0d0, c2t1d0, and c1t2d0 were all to fail, the data would still be accessible, despite the fact that all the metadevices are in a "Needs maintenance" state:

Code:

  d0: Mirror
Submirror 0: d10
         State: Needs maintenance
       Submirror 1: d20
         State: Needs maintenance
       Pass: 1
       Read option: roundrobin (default)
       Write option: parallel (default)
       Size: 7058880 blocks

   d10: Submirror of d0
       State: Needs maintenance
       Size: 7058880 blocks
       Stripe 0: (interlace: 32 blocks)
           Device     Start Block  Dbase        State Reloc Hot Spare
           c1t0d0s0          0     No     Maintenance   Yes 
           c1t1d0s0          0     No            Okay   Yes 
           c1t2d0s0          0     No     Maintenance   Yes 

   d20: Submirror of d0
       State: Needs maintenance
       Size: 7058880 blocks
       Stripe 0: (interlace: 32 blocks)
           Device     Start Block  Dbase        State Reloc Hot Spare
           c2t0d0s0          0     No            Okay   Yes 
           c2t1d0s0          0     No     Maintenance   Yes 
           c2t2d0s0          0     No            Okay   Yes

When replacing these disks, only the disk being replaced will need to be re-synchronized. Again, that is because this metadevice is being treated by DiskSuite as a RAID 1+0 device.

If corresponding disks in each submirror were to fail (such as c1t0d0s0 and c2t0d0s0), data loss would occur. as usual on a raid 1+0, it's not really saver, but it should be faster.... when you think about it, you are just working with luck, based on the probabilistic theory. the idea is that it is not presumably to lose 2 disks of the same pair, in theory. at the end of the day, only one disk should/may fail to provide a redundant lun, everything else is just a game, but think about Murphy's law:
so it's a great idea to improve the performance, but not to increase the availability

greetings PRESSY

Hi Pressy,
Thanks for your answer which is very clear as usual.
Just a precision : the solution I wanted to take for RAID 1+0 has been given by SUN support !! Problem ?

what type of storage are you using? if sun recommend you to use it, it will have a reason, what did you ask them?

a raid 1+0 is for sure the best way to build a high availability "disk" and the fastest, but to be honest with you, the most of my costumers don't use it, because it is the most expansive way, too. it costs exactly the double money of your space.... you lose 50% of your disks.... with 4 disks it doesn't matter, but think about 2TB luns so if you make a raid 5 with 10 disks, you would only lose 1 (10%), 20 disks 1 (5%), and so on and so on...
at the end of the day, it's a performance question, if you have to handle a raid 5 without a HW-controller and without HW-cache the performance is extremely bad and it costs CPU power...

i would also recommend you to use a 1+0, but it belongs to you. do you want to spend more money, speed it up and get a better availability... or save money and get more space on performance costs...

regards PRESSY

It's quite a long story.
My server has 6 disks 150 GB each but only one controller.
The first operation I did was to mount 3 disks in RAID 5 but the performances were awfull.
That's why SUN recommend me to mount the disks in RAID 0+1 or best in RAID 1+0 with the commands that I posted in my first post. The problem is that RAID 1+0 doesn't work with SVM : the last command metainit d130 1 2 d120 d121 -i 64k fails because it needs physical devices and not logical devices.
Anyway I choosed the solution RAID 0+1 and now it's fine, except that I will need a second controller to improve the performances .
Thanks again