Return to Level1Techs.com

Linux benchmarking confusion

I recently built a testbed for experimenting with different Linux storage options. One of my first tests was building an LVM RAID 5 array mounted at, oddly enough, /mnt/raid. I wanted to test the performance of this array (3x1TB 7200 RPM SATA drives.) The only Linux benchmark that came to mind was the Phoronix test suite. And I got some wild results. For example:

Flexible IO Tester 3.1:
pts/fio-1.11.2 [Type: Random Write - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 128MB - Disk Target: /mnt/raid]
Test 1 of 1
Estimated Trial Run Count: 3
Estimated Time To Completion: 2 Minutes [22:48 CDT]
Running Pre-Test Script @ 22:46:34
Started Run 1 @ 22:46:34
Running Interim Test Script @ 22:47:41
Started Run 2 @ 22:47:41
Running Interim Test Script @ 22:48:50
Started Run 3 @ 22:48:50
Running Post-Test Script @ 22:50:07
Type: Random Write - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 128MB - Disk Target: /mnt/raid:
605
575
569
Average: 583 MB/s
Deviation: 3.31%

583MB/s random write speed is kind of hard to believe from an array of 7200 RPM drives. Note that the buffered option is NO and the direct option is YES. For comparison’s sake, I ran GNOME’s disk utility benchmark with a 128MB block size and got about 180MB/s read speed and 22MB/s write speed (that seems really low, but maybe it’s right.)

Any thoughts on which benchmark is more likely to be right, or if there’s a better test out there?

Added ‘help desk’ tag so people know you need help.

OK I think I figured out what’s wrong with Phoronix. Even though I told it to test the mountpoint the RAID array was mounted at, it seems it tested the boot SSD instead. You can see it in the following test result:

Flexible IO Tester 3.1
Type: Sequential Write - IO Engine: Linux AIO - Buffered: No - Direct: Yes - Block Size: 128MB - Disk Target: /mnt/raid
MB/s > Higher Is Better
INTEL SSDSC2CT06 … 581

The Intel SSD is listed as the tested hardware instead of the array, even though /mnt/raid is specified as the target. Sigh…

1 Like

That may well be a device enumeration bug with the test suite then.

I would say write Michael. There are some bugs that he has not bothered with fixing due to how much time he devotes to other things. He may have a bounty on this already.