After reading about Nick Haslam’s tests using TPC-H data I thought I’d see how quick the new server I was building was (before it went into production!).
The server is an IBM x3650 M3 with eight disks installed: four small SSDs and four bigger SATA and SAS drives. Each of the disks is a RAID1 mirror of identical pairs.
|Type||DB Build||Data Load||Total|
|Data and log separate SSDs||0:02||02:24||02:26|
|Data and log on single SAS||0:07||02:28||02:35|
The initial test was fast but not as fast as Nick Haslam’s load to multiple files on a single C (presumably SATA) drive. I had expected to be faster but of course a bulk data load requires mainly write operations, and SSDs are notoriously slower on write operations. In addition these are running as a RAID1 array so each write requires two write operations, and the controller is set for write-through (ensure the data is written before complete).
As a comparison I reran the load with all the database files loading on the same SAS drive (a 10k rpm Savvio RAID 1 pair). As this is on the same controller it would provide a reasonable comparison of SSD vs SAS write performance.
So not as fast as writing to unmirrored disks but not massively slow either.