Thanks for both replies.
I tried to do the select top 500.000 but I'm difficulties, I will ty later.
Regarding the IO scheduler, the configuration was set to noop.
I just did a test with cfq because that was the configuration set for the production enviroment using EMC, which has a better performance on the index creation.
The cfq test present worse values (407966 ms vs 328853 ms).
Regarding the IO request queue size for the block devices, the value is set to 256.
I will change it to 1024 to test.
Regarding the filesystems, the current configuration is ext4 noatime noadirtime
Just to summarize, the current production box (physical) running EMC VNX is having in the overall better performance than the VM (VMWare) and Tintri storage.
Thanks