I am looking into adding HDD reviews to TPU ...
Does anybody here have experience how to properly benchmark drives? Please no Sandra/ATTO/HDTach unless you can explain how (cached?) read bandwidth can relate to single-/multiuser application performance in real life.
HDTach does use burst speed, and that in itself is somewhat meaningless, unless you are checking to be sure of what mode the drive is in, and possibly only to ATA drives as SATA has of yet to actually break the 3.0Gbps speed barrier.
But the disk sequential read is uncached, there are too big of write files on the "long test" to cache. CPU usage is somewhat unreliable though, as it seems to vary by as much as 2-3% per test and it even stated +/- 2%. But 2% these days isn't alot to most.
The sequential drive read speed is the most accurate way of telling about the true world performance of the drive-drive subsystem. It should directly relate to how fast the drive system is when loading game levels, as well as other disk intensive tasks.
Seek times seem important to users, however I seriously doubt the role they play in loading a game level, the data is sequential and laid out so on the disk if defragmenting is done. And as most small files are cached (windows startup files) when a system is optomised, and thumbnails are cached by windows, I see minimal difference in seek times here.
Bu to each their own and I hope that we can get conclusive results from testing done.
I could submit a few results off the different systems I admin, as well as my home system.
Here is our server, live on the network. I was-am the only active user besides network load, and it was marginal at the time.