Host Managed SSD's - This might change the reason to Over Provision (by 30% or more)...

See:
http://www.anandtech.com/show/9720/ocz-announces-first-sata-host-managed-ssd-saber-1000-hms


Basically, they are opening up the internal workings of the SSD and giving control back to the user. I.E. use the SSD for the user's workloads or, do the GC necessary to keep the drive in top condition.

What is most telling to me is that OCZ's example shows 33% of a pool of drives is left to do it's GC and TRIM... and the remaining two thirds of the drives are actively used.

Seems a lot like my OP'ing recommendations for the last few years, huh?

The difference of course is that with OP'ing, that 33% 'unallocated' capacity is sacrificed for as long as you need the high performance from the drives (usually; indefinitely).

With HMS, the full capacity of the drives are available (in-between Maintenance Mode periods), indefinitely.

One thing to keep in mind though with this first HMS enabled drive:

But whether the above disclaimer was there or not, being OCZ, these drives will never be implemented in any of my systems.

But future Intel, SanDisk or even possibly Samsung (I/S/s) offerings with HMS might be worth looking at for the many older (SATA) based systems I still use and manage.

Interesting that NVMe implementations are also possible with HMS.

That is the I/S/s models that would be most interesting to my workflows, especially for mobile systems.

Consider that today I leave many of my mobile systems (at least a dozen and usually closer to two dozen) on overnight at least twice a week or more so that TRIM and GC can happen when I don't need to use them. Desktops are left on 24/7, so they don't bother me as much for this issue.

HMS will finally give me control of when TRIM and GC happens on my storage subsystems - instead of it being a 'suggestion' to the SSD to perform TRIM/GC when it deems itself idle 'enough'.

To me, 2016 seems like the year that SSD's will have finally grown up enough to not feel like a science experiment any longer; even if to many others, that experiment seemed so successful vs. the HDD's we had grown to love/hate for well over half a century.


See:
http://www.tomsitpro.com/articles/ocz-sabre-hms-ssd-toshiba,1-2949.html

This bodes well for the throttling and overheating issues we are seeing too for higher performance drives (M.2, mSATA and even PCIe implementations).

With just 'dumb' storage on the nand based drives, the selective control and immensely more powerful CPU the host has will negate that poor aspect of currently available drives to the point where it is not an issue anymore (and... as it should have been from the beginning...).

Putting a powerful multicore processor on a storage subsystem was going backwards... yeah; it is needed for high performance. But ignoring the vastly more powerful CPU that most systems have already (properly cooled too...), was a more than a slight kludge to me. Especially when the heat/throttling aspects weren't addressed by almost all system manufacturers.

Along with Skylake, PCIe x4 SSD's and now HMS, 2016 is shaping up to bring another huge boost to my productivity.

Maybe this time, it could give us these benefits without major side effects like a hotter running system, the need to leave mobile systems on so that the GC/TRIM finish by the next day, throttling so bad (was it the heat or the GC that was the real culprit - probably both...) that the performance we actually get is only a fraction of the performance implied on the retail box.

If I had an SSD with the above capabilities today, I know how I would run it: GC/TRIM/'everything' - OFF while I was actively using the system. And once a day, turn 'everything', including GC/TRIM - ON before I needed to use it.

And yeah; I would still test with OP'ing too (up to 67%... just like I did in 2011...).

 

See:
http://www.tomshardware.com/news/ocz-charges-petersen-sec,30335.html


Woes continue for OCZ even years after being sold to Toshiba.

Toshiba must feel like crap if they believed the books they were shown...

 

Exciting news indeed, thanks for posting, tilleroftheearth! I think that SSD makers made a good decision back in the day to basically make all SSDs simple enough that they could be dropped in to any system at the time (to the OS it just looked like any other drive). However, it's quite exciting now that we're seeing a new host controller interface (NVMe) and HMS offering much more control over the drive. I think your comparison to a science experiment is perfect. We're finally starting to see standards and new features designed strictly to take advantage of SSDs.

It looks like we'll also have data that's a much more rich about the inner workings of the drive. That could be quite beneficial for looking at failures and such.

Yes, it certainly isn't all that surprising that they breaking the drive up into thirds so you have 33% OPing. That number seems to provide the best balance. It also keeps the implementation fairly simple in that they only need split the drive up into three segments. Splitting it into 10 or 20 segments might be quite a bit more challenging (truth be told, it might not be - sometimes after the first split any subsequent ones are all the same).

Hopefully 2016 does prove to be a really exciting time for SSDs.

 

Lol, how is splitting into 10-20 complicated? It's actually pretty simple, percentage wise lol... I think 33% is a bit too much.. 20% should be enough IMO...

 

I think I was pretty clear about the fact that we don't really know. What I can tell you is that splitting the drive into 3 segments (33% each) is going to be simpler (or at least no more complex) than splitting it into 5 segments (for 20%) or 10 segments (for 10%). It has nothing to do with percentages, but rather than more divisions are generally more complex.