Confusion on RAID chunk size

mfoley · 11-30-2019, 06:00 AM

I'm creating a 2TB RAID-1, which I've done before, but I'm puzzled on this one. I've created /dev/md0 as:

Code:

mdadm --create /dev/md0 --metadata=0.9 --verbose --level=1 --raid-devices=2 /dev/sda2 /dev/sdb2

When it finished, /proc/mdstat gives:

Code:

md0 : active raid1 sdb2[1] sda2[0]
      1936736256 blocks [2/2] [UU]
      bitmap 0/15 pages [0KB], 65536KB chunk

I'm about to format /dev/md0, and I've normally used:

Code:

mkfs.ext4 -v -m .1 -b 4096 -E stride=<s>,stripe-width=<w> /dev/md0

where stride <s> = chunk / block size; and stripe-width <w> = stride * 2.

However, my documentation and Internet examples show typical chunk sizes of 128KB (with some variation) whereas mdstat show mine as 65536KB, This is quite different and no examples show chunk sizes that big. I'm wondering if that's a qood number? If I use 65536 to calculate my stride and strip-size I get 16384 (assuming block size of 4KB) and 32768 respectively.

I understand that chunk size is the number of consecutive blocks written to each drive. 65536 seems like a rather large number.

These values just seem wrong. Can someone please straighten me out?

berndbausch · 12-01-2019, 03:46 AM

The man page states (typos included)

Quote:

The default when creating an array is 512KB. To ensure compatibility with earlier versions, the default when Building and array with no persistent metadata is 64KB.

Your relatively small chunk size can be explained by the old metadata version you requested.

There is also a paragraph on chunks in the md wiki.

I don't think the stride and stripe-width parameters make sense for a RAID1, which is not striped but simply mirrored (though they probably don't hurt either).