EVA6400 array sizing and performance when leveling

[Donskoy Stanislav]

I have a question about what-else Exchange Mailbox 2007 storage sizing and EVA6400 array IOPS decreased in case of EVA6400 array leveling. I use Exchange 2007 Storage Calculator and I received 55 FC disks for database LUN (and 144 disks for all system – 2 node cluster with CCR) if I use Microsoft recommended values for RAID rebuild overhead Percentage (25% for RAID10 and 50% for RAID5). RAID rebuild overhead Percentage is a value used in Exchange 2007 Storage Calculator, it determines IOPS performance decrease in case of array leveling (rebuild). MS told, that these values (25 and 50) are typical for most arrays. If I use RAID 10/5 rebuild overhead Percentage = 0, I receive 30 FC disks for database LUN (and 94 disks for all system). My question is - what are the RAID rebuild overhead Percentage values for EVA6400 array? I mean how much the number of IOPS will be decreased in case of EVA6400 array leveling? If it is possible I would like to receive some link to these data in HP documentation.

And the second question is - if I use 30 disks for database LUN, and rebuild overhead takes place (for example, IOPS decrease 50% and take 10 hours), what will happen with Exchange? If will be just a slow response for users or it queues can dramatically grow and messages will be delayed for hours?

[susanta_dutta]

Hi Donskoy,

In EVA, Leveling is not the rebuild of other storage. Leveling is just EVAâ s internal low priority house keeping process to distribute data across all the drives â Which helps EVA to improve in response time. The below customer notice has detail information whether user should worry about leveling or not?

http://h20000.www2.hp.com/bizsupport/TechSupport/Document.jsp?objectID=c01209515

Regards
Susanta

[Donskoy Stanislav]

Thank you for the answer.
As I understood when the disk fail takes place there happen two processes: at first reconstruction and then leveling. As it mentioned in customer advice, leveling process has minimal performance impact. Is it possible to count the percentage of the performance decrease? And what can be said about the reconstruction process? How long does it take and does it have some influence on the performance?

[susanta_dutta]

Hi Donskoy,

Because of virtualization concept in EVA, all the drives in the Disk Groups does the I/O for the hosts, hence the load due to reconstruction & leveling processes become negligible compared to what EVA can respond to Host I/O.

There are many factors involved in time taken in reconstruction & leveling. Number of VDisks in the disk group, Number of disks in the group, the occupancy of the disk group, host I/O load during the reconstruction & leveling and also other background activities ( creation, deletion, snapping of Vdisks etc etc ). Hence there is no straight-forward method to determine this.

Regards
Susanta