Disk Arrays
Showing results for 
Search instead for 
Did you mean: 

linear vs. disbursed data layout

ken lyles
Occasional Advisor

linear vs. disbursed data layout

Hello, I'm new to a storage mgmt team and am hearing quite a debate regarding linear and disbursed ldev layout in an Hitachi 9960. Is there a good thorough explanation of what these are and any guidlines/suggestions of how to choose one over the other. I'm an HP sysadmin with experience with EMC.....thanks
Vincent Fleming
Honored Contributor

Re: linear vs. disbursed data layout

HP lays out XP's linear. It simplifies HP-UX System Administraton, and care and feeding of the array. It also makes it simple to determine where in the array any particular LDEV is, and makes upgrades more straightforward.

The difference is when LDEVs are allocated. Linear is analgous to disk concatenation (LDEVs sequential in numbering are located on the same array group or sequentially numbered groups), where disbursed is like striping. (The LDEVs are striped onto the array groups).

With Disbursed mode, you cannot predict with any certainty where an LDEV actually is in the array. It could be anywhere.

Worse, Hitachi's tools cannot tell you where it is - you have to access the SVP to find out where it is. (HP's Command View XP actually tells you which array group an LDEV is on). Of course, you're technically not allowed to access the SVP.

The reason that some at Hitachi like to disburse the LUNs is IFF you allocate your LUNs to hosts sequentially [ie: 0:00, 0:01, 0:02, etc], then you are guaranteed to be striping across the largest number of array groups possible.

This works until you add drives to the array... if you had 10 array groups, disbursed, and add another 5 array groups at a later date and continue your linear host allocation, the performance of the new drives will seem less because you're using fewer drives in the array.

We feel that this is better done manually, because you might want to separate your logs from your tablespaces - have them on different drive sets or in different sections of the array. If you disburse the LDEVs, you will make this nearly impossible.

We also make the correlation of CU to quadrant - ie: CU 0 is always the lower right quadrant of the array. DBA's like to know where their data is, so when they ask, you can point to the drives and say, "there!".

Either way, you have to do some tracking and manual operations - Disburse means that host assignments at initial installation are easier to manage, but later on becomes ugly, and Linear means taking the ugly up front so it will be cleaner for it's lifetime.

Good luck with the array... next time I hope you'll consider buying the HP version of the array - the XP512 or XP1024.
No matter where you go, there you are.
Insu Kim
Honored Contributor

Re: linear vs. disbursed data layout

Leaner assigns the LDEV IDs in sequential order for all unassigned logical addresses within the parity group. This choice is appropriate for administartors who actively balance their disk array workload.
LVM striping among multiple gruops would be good example.

Disperse randomly rotates LDEV IDs among multiple groups across the disk array, elminating hot spots and improving data access.
This should be used by administrators that do not actively balance their disk array workload.

Hope this helps,
Never say "no" first.