>The first point to investigate is the interweaving of the lock when writing ( SQL Profile is a basic start point ).
>This depends on the database and application design, and the number of competing operations.
>
>The time can be reduced by improving the hardware,
>but the minimum value depends on the design.
>One way to find the minimum is to use ( not in production ! ) a Cache RAM (PrimoCache) that hides the disk latency.
>Then you can compare this benchmark with a RAID controller with DRAM cache and SSD storage.
Thanks, this is interesting.