I have been reviewing a fairly large SQL environment and business owners have asked the usual question - why is a much lower spec dev environment running quicker than prod. Of course there are a number of differences so no expecting the same performance but one issue that has come to my attention is that the prod environment has write latency on the tempdb of 7000ms! This compares with 3ms on the dev.
The prod environment has a tempdb on a Samsung 840 Pro 512GB SSD. I have been provided with some bench-marking performed a while ago using CrystalDiskMark which showed the throughput for this disk has been pretty poor when in RAID1 config, for example 4KQD32 writes = 77MB/s and seq writes 33MB/s. Interestingly, some investigation was done without the RAID and was getting 4KQD32 = 151MB/s and seq = 377MB/s.
If mainly used for sorting, distincts and hash joins, would the tempdb be writing sequentially to disk?
My view is that this seems like a huge hit for the redundancy it provides.
Can anyone provide any views on whether this degradation is typical of hardware RAID1 (it was not on the DELL HCL)?