TechReport Invalidates A Test

TechReport did a review of 4 intel X25-E in a RAID0 setup (A look at four X25-E Extreme SSDs in RAID – The Tech Report).

However, there were some problems with the testing method.

• RAID-0 of physical drives is dumb.  Nobody should use it, for any reason, unless the drives underneath are mirrored.  Benchmarking this is a bit like benchmarking how fast you can piss into the wind… not only do you get pee on yourself, nobody else cares.
• Windows XP’s caching algorithms make bad assumptions about disk configuration. It assumes you are using platter-based disk drives. Try the whole suite again with vista x64 sp1 (and with/without “advanced performance” mode enabled in the device manager, you should see completely different numbers.
• Also, this is a PCI-E x8 card, but it seems like it is in a PCI-E slot. PCI-E x1 = ~640MB/sec minus a little for striping (200MB/sec) = ~450MB/sec peak, which seems to be what you are seeing. I have four of those drives in a server with two separate controllers on x8 slots in RAID0 of a pair of 2-drive RAID1s and see ~500MB/sec reads with OS caching disabled: this is what I would expect with two effective drives worth, not four.

Note that, when I went to register for an account to comment there, it gave me an error, so I decided to blog about it instead.