I haven't tested the MQ policy since I'm on kernel 4.6, I am on SMQ. I have a spinning disk as the origin that is falling behind as a mysql slave. I added an ssd writeback cache but I can still see a a large amount of random writes going to the spinning disk on iotop. The filesystem is XFS. Origin is 2tb, but the dataset I am working with is only 305gb. The cache is 40gb total.
dmsetup status shows
cached: 0 3906766976 cache 8 330/1663 512 163686/163686 1781562 5956880 6792625 24895886 18847 18847 3398 1 writeback 2 migration_threshold 2048 smq 0 rw -
a minute later shows
cached: 0 3906766976 cache 8 330/1663 512 163686/163686 1781956 5959500 6794474 24908493 18850 18850 3396 1 writeback 2 migration_threshold 2048 smq 0 rw -
I was looking over the archives of dm-cache and came across https://www.redhat.com/archives/dm-devel/2015-September/msg00090.html
I understand that dm-cache does not just blindly stream all writes to the cache and then update the origin. Is there some way I can enable that though (even if it involves changing code)? The hotspot feature is impractical for this workload and is of little help it seems. I know the SSD taking writes by itself can keep up, I just can't dedicate that much SSD space to this right now. I have to turn barriers off to make the hard disk keep up and I'd rather not do that.
-- dm-devel mailing list dm-devel@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/dm-devel
