4 ms·

So as I understand the primary issue with SMR drives is that non-sequential writes tends to be slow because several tracks needs to be rewritten. Wouldn't the p

So as I understand the primary issue with SMR drives is that non-sequential writes tends to be slow because several tracks needs to be rewritten. Wouldn't the perfect solution be to make it a hybrid disk instead (i.e. add a small SSD to it)? That way the data can be written to SSD quick and then the firmware can move it onto the platters slowly. Of course there are some issues about balancing SSD size vs. cost and write cycles, but at least if designed properly the failure mode once the SSD part is weared down would be that it just functions as a regular (device managed) SMR drive.

jamilbk 9y ago
This can essentially be accomplished with ZFS and an SSD as a L2ARC device. I have three 8 TB Seagate SMR drives in RAID-Z (aka RAID-5) with a 500 GB Samsung 950 Pro M.2 as a L2ARC cache drive on the pool and it works beautifully for my workload. In my experience, once the scratch area of the SMR drive is full, I get about 4-5 MB/s sustained write speed, which in my pool translates to 10 MB/s for the pool. Since the scratch area is about 25 GB, that means I can do 500 GB + 40-50GB of random writes before things slow to a crawl, and I have to wait (550 GB / 10 MB/s) = 55,000 s ~ 16 hours for the writes to flush. I paid $179.99 each for these. Not Too Shabby.