http://www.zdnet.com/article/why-ssds-are-obsolete/

 

I enjoyed this article. Made me think about why SSD's exist and what could / should be done when the engineers apply some further research and experiment.

This guy is clueless.  Moving from AHCI to NVMe is the next step, not moving the FTL to the OS.

There are already PCIe M2's that interface with 4 lanes (PCIe x4) ... and add-on cards that provide PCIe x8 interfaces.

 

These are FAR faster than a SATA-III interface.

 

Of course, they have price tags that aren't likely to make them mainstream storage anytime soon -- and to use one of the PCIe interfaced M2's your motherboard has to have the appropriate connections.

 

I think, however, it will be quite a while before SSDs are actually "obsolete" in any real sense.  In fact, as the price continues to drop, I suspect they'll be much MORE prevalent ... not less.

 

Yes, but make sure your M.2 slots are PCIe 3.0, not 2.0.  Also make sure your lanes are not being diverted (some mobos share the lanes between M.2 slot and a PCIe slot), and you have enough lanes to start with.  With dual PCIe x 8 hard drive controllers, plus another PCIe x 8 for 10GBit Intel card, the M.2 slot SSD will bring you to 28 lanes, and that is before counting the rest of the rig such as video.

 

I'm looking at a new mobo with a 40-lane Haswell E, and waiting for the Mushkin Hyperion to come on the market.  I was considering the Samsung XP951, but they decided to abandon NVMe and it is only AHCI.

There are already PCIe M2's that interface with 4 lanes (PCIe x4) ... and add-on cards that provide PCIe x8 interfaces.

 

These are FAR faster than a SATA-III interface.

 

Of course, they have price tags that aren't likely to make them mainstream storage anytime soon -- and to use one of the PCIe interfaced M2's your motherboard has to have the appropriate connections.

 

I think, however, it will be quite a while before SSDs are actually "obsolete" in any real sense.  In fact, as the price continues to drop, I suspect they'll be much MORE prevalent ... not less.

Which is actually Robin's point, in his badly titled article. Unfortunately both him and the Phd candidate he tried to source overlooked the work already done, by Samsung no less, f2fs. Architectural, the indirection layers are key to allow the advances in each layer, without requiring so much change as to stall the advance. Bad Robin, bad. Might as well try to say RAID5 is dead. He has points, but when searching for a title he tends goes too far allowing many to dismiss his regurgitations.

 

After all how do you find a flash m.2 device? Better look in the SSD section 

 

... After all how do you find a flash m.2 device? Better look in the SSD section 

 

Indeed 

Perhaps he should have titled his article "Why SATA SSDs are obsolete" 

 

After all how do you find a flash m.2 device? Better look in the SSD section 

 

Of course.  It is a Solid State Drive. 

 

The list is kept here:  http://www.johnnylucky.org/data-storage/ssd-database.html

The problem now is not accessing the flash itself, even through FTL, but the overhead of the storage protocol (AHCI) in handling the I/O requests.  Just like SATA III has a top bandwidth of 6gbps, no matter if you have 0 latency flash, there is still latency in the AHCI overhead.  That is the main benefit of NVMe over AHCI.

 

The FTL is an abstraction layer, but it is highly tuned with its on dedicated controller.  Eliminating it will no make a lot of difference in real world practice --- but moving from AHCI to NVMe will, particularly in smaller I/O which is impacted the most by latency.