Western Digital's Ultrastar DC ZN540 Is the World's First ZNS SSD

An anonymous reader quotes a report from Tom's Hardware: Western Digital is one of the most vocal proponents of the Zoned Namespaces (ZNS) storage initiative, so it is not surprising that the company this week became the first SSD maker to start sampling of a ZNS SSD. When used properly, the Ultrastar DC ZN540 drive can replace up to four conventional SSDs, provide higher performance and improve quality of service (QoS).

ZNS SSDs have a number of advantages over traditional block-based SSDs. For one, they place data sequentially into zones and have better control over write amplification, since the software 'knows' what it is dealing with. This means that ZNS SSDs don't need as much overprovisioning as traditional enterprise drives. Many enterprise drives rated for 3DWPD (drive writes per day) reserve up to 28% of their raw capacity for overprovisioning. ZNS needing as little as a tenth of that significantly increases usable SSD capacity. Second, since ZNS manages large zones rather than a bunch of 4KB blocks and doesn't need to perform garbage collection as often as traditional SSDs, it also improves real-world read and write performance. Finally, ZNS substantially reduces DRAM requirements.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[ArchieBunker]

Ok that made me lol

Re: Talk dirty to me

[MrNaz]

Pick the guy who never worked in IT.

Re:Talk dirty to me

[sjames]

Because the world is a perfect place where UPSes and power supplies are 110% reliable and nobody, not even a 5 year old girl, ever presses the emergency power off button or unplugs the wrong power cable.

Of course, every home user has a UPS with perfect predictive failure and qualified personnel to test the battery weekly.

Re:

[bill_mcgonigle]

The comments above say they're getting /rid/ of the complex abstraction layer. So if your journal is corrupt you discard it. Hopefully your app used fsync if it needed to.

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[WorBlux]

Depends of the drive. Most enterprise drive engage in cooperative wear leveling were the disk signals a zone is about to wear out, the OS catches it and schedules data movement as a low priority task. When the zone is empty then the disk is free to retire it or swap out the mapping to solve the issue. And even in the case of a potential consumer drive that doesn't make latency and predictable access times priority number one, you don't have to operate your algorithm on the 512B level, and likely using much

Re:Talk dirty to me

[thegarbz]

So how do these drives handle a dirty shutdown

The same way every other journal based system handles it in software, or current SSD handles it in hardware. Discard the most recent entry, and put enough capacitors in the drive to allow the current write to finish.

This isn't adding a layer, it's removing one by getting an SSD to stop pretending its something it's not. It literally just exposes a portion of the NVMe command set to the host OS directly and stops saying "hey I'm a 1980s era HDD, trust me, *wink*" like the current SSDs do.

It's a journaling filesystem done in firmware

[arglebargle_xiv]

For those who have never heard of ZNS, it's something like a journaling FS done partly in drive firmware, but with extra support needed in the OS to handle it. As opposed to a standard journaling FS supported in the OS.

I have no idea what the point of such a thing is, given that you can get a full-blown journaling FS without ever needing to buy a ZNS drive for it.

Re:It's a journaling filesystem done in firmware

[PhrostyMcByte]

The problem is that SSDs already behave like this, but also have a complex middle layer that hides it from the OS and exposes everything as purely a random-access block device.

This removes that middle layer with the idea that a supporting filesystem will have more knowledge about the writes it performs, allowing improved performance and cheaper drives. They've basically built a drive that only works with journaling filesystems.

Re:

[enriquevagu]

Thanks!

Re: It's a journaling filesystem done in firmware

[sectokia]

Your post is garbage. Itâ(TM)s nothing to do with journaling. SSDs are lba block devices which means they have to internal accept read and write for any lba at any time randomly. This is hard to do in practice because flash memory needs to be erased and even written in blocks much bigger than an lba. Zoning allows you to specific ranges of LBAs so the drive knows how many writes to expect maximum in that zone regardless of sequential or random. You open a zone, make whatever writes and close the zone

Re:

[gweihir]

I have no idea what the point of such a thing is, given that you can get a full-blown journaling FS without ever needing to buy a ZNS drive for it.

The point is selling drives to the clueless. Corporate IT is often full of them, especially the "decision makers".

Re:It's a journaling filesystem done in firmware

[thegarbz]

I have no idea what the point of such a thing is

With an explanation like the one you just gave this goes without saying. I can imagine anyone who thinks its a Journal FS done in firmware would not understand the point, because there is no point.

Fortunately it's nothing like that at all.

What it *is* is exposing the SSD for what it is rather than the usual method of having an SSD pretend to be an old HDD. ZNS is literally just exposing a chunk of the NVMe command set directly to the OS and letting the OS address it directly.

Re:

[arglebargle_xiv]

So it's acting as an MTD? That's why I assumed ZNS was doing something more than just "expose the raw interface", meaning "act as an MTD".

Re:

[vyvepe]

ZNS has nothing to do with journaling except that journaling file systems tend to write more sequentially which is useful for zoned storage devices.

Zoned storage is actually exposing underlying flash more directly (without a middle layer implementing logical block address abstraction of a classical disk above a flash storage). The basic commands possible on a zone are zone erase (corresponds to flash bank erase) and sequential block write to a zone (corresponds to writing to a flash). Before writing to a fl

Re:

[account_deleted]

Comment removed based on user account deletion

Re:

[WorBlux]

sequential write is a requirement for zns, it's just that the drive can the sequencing for youin certain cases, via the zone append command. Otherwise is works a lot like zbc/zac for SMR drives.

Re:

[labawi]

Maybe you meant a COW filesystem? Common SSDs have FTLs to remap blocks, because they can't overwrite random block at any time, much like a COW filesystem does not overwrite data.

FTL - flash translation layer

Re:

[WorBlux]

The point of it is that the underlying flash technology is inherently sequential. By guaranteeing the use patterns will more closely match the physical constraints of the memory, you can make the drive simpler, cheaper, and more reliable and predictable.

the software 'knows'

[QuietLagoon]

Anytime I see that nowadays, I have to ask.... how much of what the software knows is the software sharing with data harvesting entities?

A drive with built in RAID

[rapjr]

I've often wondered why hard drive makers don't offer the option of enabling a built in RAID on their drives, letting you cut your drive size in half while offering greater reliability. Seems like it would be simple to do and be a marketing differentiator, letting people easily use RAID without the hassle of buying a special controller and configuring it. The technology described in this article suggests it is possible.