Hi all, finally I had some time to check the performances, albeit in a very simple way, between a RAID-10 near 2 and a RAID-1. The setup had 2 SATA SSD and 2 NVME SSD, delivering two different storage configurations (2 SATA or 2 NVME). The test was a simple sequential read from the raw device (/dev/mdX), with read block size of 1MB. The RAID-10N2 had chunk size of 512KB (default), which happens to be exactly 1/2 of the read block size (not by accident). This means it would be optimal to read from two devices, in case of RAID-10N2. For the RAID-10N2 and the SATA SSDs, I got ~920MB/s transfer rate, not so stable, but reproducible. For tha RAID-1 and the SATA SSDs, I got ~480MB/s, which is in line with the single SSD (raw) device sequential read. For the RAID-10N2 and the NVME SSDs, I got 3390MB/s, extremely stable and reproducible. For the RAID-1 and the NVME SSDs, I got 2490MB/s, still extremely stable and reproducible. In my view it is pretty clear that the RAID-10N2 reads interleaving the two devices (visible by "iostat -m 5), and saturates whatever bottleneck is there. I can imagine that, with HDDs, this will bring nothing, since the head of the HDD has anyway to fly over the skipped blocks. In SSDs, on the other hand, this is not an issue. For the RAID-1, is also clear that one device is involved in the transfer (not so visible with "iostat -m 5). Of course, as stated at the beginning, this is a very very simple test, it does not take into account anything else except raw sequential read performances. Nevertheless, this was my initial doubt. Hope this helps, bye -- piergiorgio
