Originally posted by: stephen2615
I come from a UNIX background with Veritas and never had any concerns about performance but now that I have Windows hosts that don't have any volume management, I would like some advice.
I have a request for nearly one TB of storage for a system that runs SAP probably with SQL Server. I have a number of parity groups available that are just under 1 TB (7+1 Raid 5 with 146 GB 15 RPM disks). With a volume manager, I would have carved up the PGs across a number of BEDs and stripped them to get the 1 TB LUN. The storage is a USP.
The way my employer currently presents storage with the Windows systems is with smallish LDEV's used to create a LUSE LUN. Someone suggested 10 x 100 GB LDEVs made into the 1 TB LUSE LUN. I am not all that convinced that that is a good idea as I have seen major problems before with large LUSE LUNS especially as the data starts filling up the entire LUN. Reads in particular get bad as more data fills the LUN. The thought behind the 10 LDEVs was that the LUSE LUN would be across 70 disks.
I suggested that the Windows server would be better off using the 10 LDEVs and letting SQL Server sort out the database issues but that has not been accepted all that well.
Would presenting one 950 odd GB LUN from one PG have worse or better performance that the option presented above.
Has anyone got any suggestions (other than to use a volume manager) to get the best performance for the database. As it is for SAP, I imagine it might be hammered a bit. If the only option is to use a volume manager, then perhaps it is time to think about it for large capacity systems.
True copy and/or Shadow image are probably a possibility once this gets into production.
Any assistance would be most welcome.