Hi Techs

I have question based on the best practices to be followed while deploying SQL/Oracle. Let me give you a brief of our environment. We have VNX7500 with fast cache (No FastVP). traditional/Pool luns (no thin luns) with Cisco MDS9500 directors and the Frontend is ESX4.x/5.x. We have approx 1500 VMs and approx 100 SQL servers with 20 odd oracle machines. My question is specifically on storage luns layout when assigning luns for these two databases VMs. Let me give an example of a scenario which generally comes across i.e. 500 GB lun requirement for building a VM for SQL server and the following are the partitions. OS partition 50GB, Data 150 GB, Logs 50 GB. Archive logs, 50 GB & tempDB 50 GB.

Now in the above scenario I am aware if i assign a single 500 GB from any pool or RG it causes drive contention issues since the logs/tempdb/DB/Archiving logs are processed by the same set of spindles. So we follow a thumbrule of ensuring we assign 4 different luns from four different RGs and pools to avoid any contention issues as per the EMC white papers. Now the luns we create will be as per the size of the original requirement for each SQL component. Now the challenge we face is due to the high amount of SQL servers i am quickly filling up the powerpath requirement of supporting max 255 luns on a host.

Now heres my question

What I plan to do is instead of assigning smaller luns i assign a 500 GB luns for each component i.e. DATA+OS drive on one lun, Logs on the 2nd Lun, archive log on the 3rd and Tempdb on the 4th. Now since my requirement is less but I am assigning 4 x 500 GB luns can I use the free space for similar other requirement of SQL ensuring i use the naming convention of DATA/Logs/Tempdb/Arch while creating VMFS partition and any new SQL server built I use the Database lun for Data, Logs lun for log and so forth for the 2nd VM if requirement arises? (We are using Mirrorview/A for DR replication and put these luns in consistency group).

If I stick to my 1st approach which I am sure no one would disagree and its the best approach but it has a bottleneck of assigning multiple luns for a single SQL VM.

If I go with a second approach I fear of data inconsistency for multiple DB Vms on similar luns which are shared by other SQL VMs and the replication is happening through Mirrorview/A with the luns in consistency group.

Any new way of solving this issue or any general practice you guys follow will be very helpful. I am sure many would have come across with a similar situation but not sure if you compromise on disk contentions or you stick to the 1st approach.

regards

Firoz

+44-7919381259