Testing sequential cutoff

Hi,

what is the correct way of testing sequential cutoff?

I've:
- writeback mode
- sequential_cutoff to 1M

I'm doning:
dd if=/dev/zero of=/dev/bcache0 oflag=direct bs=16M count=200

After that dirty value is at 2.3G and bypassed is 29M.

I tested this with 3.17-rc4 and 3.10 kernel same results.

What's wrong here?

Thanks!

Greets,
Stefan