Greetings,
I've tried to max out the sequential read speeds of 4 x NVMe drives in Raid 0 configuration. The setup is exactly same than seen in here: https://www.youtube.com/watch?v=5wICTa4I-lg
If I read in parallel directly from 4 devices I have combined read speed comparable to the Windows test. However, when any kind of file system is introduced then I cannot get read speeds above 6GB/s. In fact this happens even if I do the same test with RAM disk. Is there some bottleneck in Linux file system handling or am I just using too old kernel? The system is running on CentOs 7 and kernel 3.10. I've used hdparm and 4 parallel DDs for the test.
I've tried to max out the sequential read speeds of 4 x NVMe drives in Raid 0 configuration. The setup is exactly same than seen in here: https://www.youtube.com/watch?v=5wICTa4I-lg
If I read in parallel directly from 4 devices I have combined read speed comparable to the Windows test. However, when any kind of file system is introduced then I cannot get read speeds above 6GB/s. In fact this happens even if I do the same test with RAM disk. Is there some bottleneck in Linux file system handling or am I just using too old kernel? The system is running on CentOs 7 and kernel 3.10. I've used hdparm and 4 parallel DDs for the test.