That’s good to know. But I think ZIL will help reduce the random writes by doing some bundling and additionally it’s effectively 1/4th the write traffic per HDD because of the array you are using. That’s probably enough to overcome the issue with current traffic levels.
When used as a separate drive, the experience would likely be very different as @fry already mentions.
For what it’s worth Seagate says NAS and SMR don’t mix. So their Ironwolf lines will not use the tech. Of course that doesn’t excuse the quiet use of it in desktop models.
And yeah, each drive gets less than that due to being in a 6 drive array. I have also set zfs_txg_timeout to 15 seconds instead of the default 5, maybe that helps too. I changed it to 30 now, will see if there is any difference.
The bigger fear with SMR in an array like this isn’t day-to-day use, since theoretically the workload written to any one drive could be bursty enough to not trip up SMR. It’s a rebuild, when you’re going to be doing a huge amount of writes to that disk, enough to exhaust the non-SMR write caches.
if my math are correct :
taking 25Mb/s => 90 000 Mb/h => 11 250 MB/h => 11.25 GB/h => 270 GB/day
depending on the way the raid / filesystem you might be below the 150GB/day I said previously, but you might be close :s
Oh, yeah, agreed. This stealth-deployment of SMR in a market segment where that matters is absolute garbage, and I’m keeping SMR out of my array as long as I can.
This seems like the kind of thing where zfs could theoretically work around it, if the drives (and drive manufacturers) were honest about how they operate. But we’re not there today.
In theory host-managed/host-aware SMRs should work just fine with filesystems aware of SMR/zones. Seagate even did make some experiments in this regard. The problem is, there are no production-ready filesystems that would be aware of zones. If/when this happen, then SMRs will probably be a decent choice. But not earlier than that.
There is a direct quote from Seagate saying both Ironwolf and Ironwolf Pro are SMR free in the article I linked.
Seagate confirms that we do not utilize Shingled Magnetic Recording technology (SMR) in any IronWolf or IronWolf Pro drives—purpose-built for NAS solutions. Seagate always recommends to use the right drive for the right application.
And while it apparenty they would hide the use of SMR in some drives, I seriously doubt they would outright lie about it.
It’s not that clear cut. Since these drives have a decent amount of cache as well as PMR areas on the disk to buffer quite a few unshingled writes. You’ll only notice it with sustained writes for quite a long time.
Blacks are still desktop drives. They make a point to say it’s for gamers. I’m sure it performs well as long as you don’t have long sustained loads. Worst case for gamers would be a download of a large game over a very fast connection. That’s at most like 100GB if you also get 4K texture packs or something. Usually much smaller. Who knows, maybe it has a 100GB PMR segment to account for that.
For what it’s worth Storj is a pretty fringe use case. For most other uses it’s probably barely noticeable. I’m more concerned with the Red’s which are likely used in RAID5 and RAID6 arrays which can have long rebuild times. So those really shouldn’t run into SMR limitations.
That said, non of these should probably be bought if you’re planning to run a node on it.