Windows – RAID 1 not performing as expected

performanceraid-1windows 7

I recently bought a new 320Gb hard drive for my computer to set up RAID 1 on it for some added security. Installation went as smooth as could possibly be (plug in power, plug in data cable, start up computer, Intel software recognized new drive, right click create RAID 1, done!).

However, for some inexplicable reason, I seem to have strange test results when using BENCH32. On my old configuration, a single 7200 RPM drive, I achieved about 60 MB/s write and 70 MB/s read. With a new RAID 1 configuration, I would expect the write to be slightly diminished but read to be significantly improved (though not exactly double speed). However, with the new configuration, I am getting 90 MB/s write and only about 80 MB/s read.

I should NOT be getting improved write performance, especially NOT better than read!

What's going on?

My system setup is:

q6600 2.4ghz CPU
4Gb DDR2 667mhz RAM
on board Intel ICH9R "RAID chip"
2x Seagate 7200 RPM 320GB drives in RAID 1
Windows 7 home premium 64-bit

Best Answer

Windows 7 (and friends) seem to make use of some fairly aggressive software write caching, which on my Windows Server 2008 R2 workstation (using two RAID-1'd drives), causes all sorts of wild memory consumption fluctuation during large sequential I/O operations. This tends to skew drive write speed benchmarks a fair bit, so you might want to turn it off for before/after comparisons.

To do so, you would go to Control Panel -> Device Manager, expand the Disk drives column, go to the Properties for your RAID-1 volume, and under the Policies tab, disable both write caching options. (The second should not be set on your onboard Intel controller unless you've got a UPS, and even then I wouldn't trust it.) Rerun your benchmarks and see if your results are as odd.

Aside from that, have you benchmarked just the new drive on it's own? I have seen instances where a slow drive and a fast drive RAID'd will "meet in the middle" performance-wise; your new drive may be quite a bit faster than your old one, and the latter bottlenecking the former read-speed wise. (Intel's software-y fakeraid is almost definitely prone to this; dedicated hardware controllers will probably handle split reads of dissimilar-performing drives better.)