4k IOPS or what? is the best benchmark for real world performance

Not open for further replies.


Jul 10, 2011
I like to use hd tune to benchmark my drives and have read about 4k IOPS being a good measure of real world performance.

is that just for ssd and not valid for regular hard drives?

I just set up raid 0 with three 500gb drives and its seems pretty fast but it only gets 115 4k iops.

compared to a vertex plus ( the cheapest 60gb drive ) gets about 6700 4k iops
and my sata 3 corsair force 3 is getting 4600 4k iops ( interesting that its less then the vertex plus)

any way there is no possible way that the raid setup i just made is 60 times slower then my vertex
it seems quite snappy in my limited time on it I am pretty happy with it.
so what should i be looking at????

I am open to a different benchmarking tool but also just a different number on hd tune


Oct 10, 2011

the 4k ios refers to access times of the drive and yes those numbers are most likely correct. think physical drive with something like 8.9ms or 12ms while a ssd has something like .1 ms
4K IOPS means "number of I/Os per second for requests of 4KBytes of data". 4KBytes of data isn't very much, so transfer rate isn't really a bottleneck in this kind of test. What the test is really measuring is random access performance.

SSDs have access times that are about 100X faster than a hard drive (hard drives = around 8ms for a decent one, SSDs = around .08ms (80 microseconds). This, of course, is exactly the reason you pay extra for each byte of SSD storage.

RAID does not improve access times. RAID CAN improve IOPS performance because you can have multiple drives doing I/O simultaneously, but for any one given I/O request the RAID set can't return data any faster than a plain hard drive.

So random I/O performance can be somewhat better (than non-raid individual HDDs) if your workload does parallel I/O, but if it doesn't then don't expect much of an improvement. And you're never going to get anywhere near the huge gains you see with SSDs.
100X faster access times don't translate directly into 100X more IO/s per second, any more than doubling the horsepower in a car will cut it's 0-60 acceleration times in half.

As to what's the "best" measurement of performance, it depends entirely on your workload. If you spend most of your time copying large files around then random I/O performance is almost completely irrelevant. What's representative for one person probably isn't for the next guy.
Not open for further replies.