SMR Drives: Are they too late to the game?

The sudden popularity over NAND Flash has spelled doom for traditional magnetic Hard Disk Drives (HDD). For years we have been hearing how HDDs are reaching the end of their life. We have also heard the same about Tape drives, long before that. Although, it would seem that the prediction on HDDs may become a bit more of reality, sooner than expected.

Recently, DRAMExchange, analyzing past trends, predicted that in the consumer market alone, Solid State Drives (SSD) will ship with 42% of all laptop sales by 2017. 2015 has seen 26% while 2014, 21%. As NAND gets more condensed into a 2.5″ (and sometimes smaller) form factor, capacities are beginning to catch up to their slower spinning counterpart. DRAMExchange also predicts that by 2017, the dollar per Gigabye of NAND Flash will drop to $0.17 while HDDs will maintain their $0.06, making the choice between both technologies a no brainer.

What are driving these prices down? Simple answer: manufacturing. As more and more SSDs are manufactured and shipped, production costs decrease. It also helps that the promise of 3D NAND will enable greater capacities than the traditional 2D Multi-Level Cell (MLC) NAND technology. This alone could potentially contribute to price drops of 2D NAND as vendors try to get them out of the door.

Also in a recent announcement Samsung announced the 16TB 2.5″ SSD. Impressive, right? To achieve such density, Samsung crammed 48 layers of 3 bits-per-cell Triple Level Cell (TLC) 3D V-NAND Flash into a single die. To put this into perspective, this can be compared to 24 layers in 2013 and 36 layers in 2014. While initial pricing of this item may be around $5,000, I can only imagine that for the same reason highlighted above, traditional MLC prices will drop further.

You may be thinking to yourself, why would anyone pay $5,000 for a 16TB SSD when they can pay $550 for an HGST Ultrastar He10 10TB Helium HDD? The answer: IOPS. Another acceptable answer would have been: zero latencies of random I/O.

Let us look at the other end of the argument, that is, archival. This is your environment that will predominantly see sequential (mostly write) I/O. Read I/O usually occurs during disaster recovery. This is where IOPS do not matter. Sure, it would be foolish to invest in such an expensive storage medium for archiving cold data. Recent breakthroughs have breathed in new life into the age-old Tape technology. IBM, the father and pioneer of tape technology, has been experimenting with a new technology that is able to squeeze 123 Gbit per square inch of tape, offering cartridges reaching as high as 220TB. That is 61x greater than the density of current LTO-6 tape. Wow! Realistically, Tape continues to make for an excellent backup medium and with a 220TB cartridge, it does make it that much more attractive.

With the increase of SSD capacity at reducing costs and with an increase in Tape capacities, where does that leave the HDD? The answer: irrelevant, at least for the next couple of years. In development for a few years now is the Heat Assisted Magnetic Recording (HAMR) technology. Initially shipping in lower Terabyte volumes, companies which include Seagate are promising 20-25TB by 2018.

What is HAMR? It is a laser technology capable of writing smaller and more stable bits onto the magnetic surface of a spinning disk. While HAMR continues to be in development, these same vendors (really just Seagate and HGST) have produced a sort of in-between technology called Shingled Magnetic Recording (SMR) disk drives. This technology too promises higher capacities but cripples under random I/O access. Three separate versions of this drive technology are offered: (1) Drive Managed, (2) Host-Aware, and (3) Host-Managed. Shingled recording writes new tracks that overlap parts of the previously written track. and because of this shingled architecture, it becomes necessary to separate tracks into zones. This is where the drive types begin to matter. In a Drive Managed SMR, the whole concept of zone and zone management are hidden from the user. All of that magic occurs underneath the hood. The same is also said of the Host-Aware SMRs, the only difference being that a few SMR specific commands are exposed for the host Operating System managing these drives. The last, Host-Managed, is being offered by HGST alone and is essentially an open box. The user needs to provide an application or write code to access these drives; that is, not going through the traditional block device subsystem of the OS. That is because the Host-Managed will report to the Operating System’s SCSI subsystem as a Peripheral Device type of 14h (Host Managed Zoned Block Device), while the other two will continue reporting as a type of 00h (Direct-access block device). What this translates to is: the latter will enumerate as a normal block device while the former will not.

Now, if HAMR is set to become more available by 2017 to 2018, why SMR? You can purchase 10TB SMR HelioSeal HDDs from HGST. Not to be confused with the Ultrastar He10 drives. A whopping 10TB that is already available. The truth is, when HAMR becomes more available it will already trump the capacities SMRs reach by then. So, I ask you, what is the relevance of SMR drives? It seems more like a throwaway stepping stone until HAMR is unleashed to the world.

One thought on “SMR Drives: Are they too late to the game?

Comments are closed.