The five-second measurement duration time limit was added to ADM2.5 in order to avoid the impractically lengthy 4K read/write measurement time (many minutes and hours) on inherently slow hard drives, to avoid unnecessary writes on SSDs, and to be consistent with CrystalDiskMark, you may get different or inconsistent random 4K write speed results.
The random 4K write result could be faster, slower, or about the same with the five-second limit. I suspect that this is due to how the write cache on the SSD device itself behaves.
- Faster: It is possible that the write cache was used during the five-second measurement. Not actually written to the flash memory.
- Slower: It is possible that the write cache didn't kick in during the five-second measurement.
- Same: It is possible that the write cache was used at the same rate during the first five-second and after the first five-second, or there is no significant write cache on the device.
Code: Select all
$ iostat -w1 disk1
disk1 cpu load average
KB/t tps MB/s us sy id 1m 5m 15m
0.00 0 0.00 26 4 70 1.82 1.58 1.47
4.02 12425 48.73 25 36 38 1.82 1.58 1.47
4.00 8507 33.23 28 41 31 1.82 1.58 1.47
4.00 6436 25.14 24 36 39 1.82 1.58 1.47
4.00 4812 18.80 35 38 28 1.82 1.58 1.47
4.00 4331 16.92 38 37 25 1.99 1.62 1.49
4.14 1973 7.97 31 20 49 1.99 1.62 1.49
0.00 0 0.00 20 4 76 1.99 1.62 1.49Code: Select all
$ iostat -w1 disk0
disk0 cpu load average
KB/t tps MB/s us sy id 1m 5m 15m
0.00 0 0.00 0 0 99 1.24 1.21 1.45
4.00 10397 40.62 3 23 74 1.24 1.21 1.45
4.00 12858 50.25 1 35 64 1.30 1.22 1.46
4.00 9139 35.70 1 33 67 1.30 1.22 1.46
4.00 7446 29.09 1 32 67 1.30 1.22 1.46
4.00 6754 26.38 0 31 69 1.30 1.22 1.46
6.96 6050 41.12 0 27 73 1.30 1.22 1.46
10.28 6418 64.41 2 9 89 1.28 1.21 1.45
0.00 0 0.00 0 13 87 1.28 1.21 1.45