Unaligned writes on block devices

I’ve read that write of block of unaligned size can induce extra reads.I mean writes on block devices in Linux. Why? How can I see it?

Answer

Block devices can only be written or read at their native block alignment, which for everything I’ve ever encountered is either 512 bytes or 4096 bytes. You can see what your device’s block size is in sysfs:

# cat /sys/block/sdb/queue/logical_block_size
512

Why? The NVMe, SCSI, and ATA command sets simply don’t support accessing a region smaller than that. The arguments to the WRITE command are in integer blocks.

If an application needs to write a smaller, or a non-aligned region, then the kernel issues a read to fill in the gaps, then writes out a larger chunk. For example, say you needed to write 256 bytes in the middle of a 512-byte block. The kernel would read the entire 512 bytes from disk, merge your data to write in the correct place, and then write the 512 byte block.

One easy way to monitor reads and writes on your system is with the iostat utility, which is packaged with the sysstat package at least on Centos/RHEL.

[root@bb-cluster-4 md]# iostat -xyz 1
...

avg-cpu:  %user   %nice %system %iowait  %steal   %idle
          32.29    0.00    7.61    0.00    0.00   60.10

Device:         rrqm/s   wrqm/s     r/s     w/s    rkB/s    wkB/s avgrq-sz avgqu-sz   await r_await w_await  svctm  %util
sdd               0.00     0.00    1.00    0.00     4.00     0.00     8.00     0.00    1.00    1.00    0.00   1.00   0.10