cancel
Showing results for 
Search instead for 
Did you mean: 

A large number of small luns versus a single large lun on NetApp with VxVM

Hi experts,

     As i known, if would like to perform better performance then we must create VxVM as stripe layout to spread all IO across disks or luns but in case of NetApp storage array if luns configured under same aggregate volume then IO will automatically distribute to all disks in aggregate volume instead, so i want to know what recommend configuration if we use VxVM on NetApp storage? or any best practice? use the single large lun size (distribute IO by Storage array) ?or number of small luns (distribute IO by VxVM first then do again by Storage) ? 

 

zerog

1 Solution

Accepted Solutions
Accepted Solution!

I didn't found any best

I didn't found any best practice document for this, but I would assume it is best if we simply let storage to distribute the IO ..

Reason being, an a normal striped volume (on physical disk), different stripelets are made to distribute IO so we get different spindles & multiple heads (of hard drives) spinning for faster access however in your case we don't have physical disks as such, its again a logical volume presented to volume manager from Netapp.. so if you cut stripelets from this logical volumes, ultimately the spindles are same in count (same spindles as in logical volume) ..

 

so to my understanding we are  not adding any value by distributing the IO twice .. just my opinion...

 

G

View solution in original post

1 Reply
Accepted Solution!

I didn't found any best

I didn't found any best practice document for this, but I would assume it is best if we simply let storage to distribute the IO ..

Reason being, an a normal striped volume (on physical disk), different stripelets are made to distribute IO so we get different spindles & multiple heads (of hard drives) spinning for faster access however in your case we don't have physical disks as such, its again a logical volume presented to volume manager from Netapp.. so if you cut stripelets from this logical volumes, ultimately the spindles are same in count (same spindles as in logical volume) ..

 

so to my understanding we are  not adding any value by distributing the IO twice .. just my opinion...

 

G

View solution in original post