NVMe flash storage doesn't mean tape and disk are dying
The advent of NVMe, software-defined storage and other newer storage technologies doesn't mean we'll be looking at all-silicon-based storage any time soon.
Not long ago, a major hardware vendor invited me to participate in a group chat where we would explore the case for flash storage and software-defined storage. On the list of questions sent in advance was that burning issue: Has flash killed disk? Against my better judgment, I accepted the offer. Opinions being elbows, I figured I had a couple to contribute.
I joined a couple of notable commentators from the vendor's staff and the analyst community, who I presumed would echo the talking points of their client like overzealous high school cheerleaders. I wasn't wrong.
Shortly after it started, I found myself drifting from the nonvolatile memory express (NVMe) flash storage party line. I also noted that software-defined storage (SDS) futures weren't high and to the right in the companies I was visiting, despite projections by one analyst of 30%-plus growth rates over the next couple years. Serious work remained to be done to improve the predictability, manageability and orchestration of software-defined and hyper-converged storage, I said, and the SDS stack itself needed to be rethought to determine whether the right services were being centralized.
Yesterday's silicon tomorrow
I also took issue with the all-silicon advocates, stating my view that NVMe flash storage might just be "yesterday's silicon storage technology tomorrow," or at least a technology in search of a workload. I wondered aloud whether NVMe -- that the "shiny new thing" -- mightn't be usurped shortly by capacitor-backed dynamic RAM (DRAM) that's significantly less expensive and faster. DRAM also has much lower latency than NVMe flash storage because it's directly connected to the memory channel rather than the PCI bus or a SAS or SATA controller.
The vendor tried to steer me back into the fold, saying "Of course, you need the right tool for the right job." Truer words were never spoken. I replied that silicon storage was part of a storage ecosystem that would be needed in its entirety if we were to store the zettabytes of data coming our way. The vendor liked this response since the company had a deep bench of storage offerings that included disk and tape.
I then took the opportunity to further press the notion that disk isn't dead any more than tape is dead, despite increasing claims to the contrary. (I didn't share a still developing story around a new type of disk with a new form factor and new data placement strategy that could buy even more runway for that technology. For now, I am sworn to secrecy, but once the developers give the nod, readers of this column will be the first to know.)
I did get some pushback from analysts about tape, which they saw as completely obsoleted in the next generation, all-silicon data center. I could have pushed them over to Quantum Corp. for another view.
The back story
A few columns back, I wrote something about Quantum exiting the tape space based on erroneous information from a recently released employee. I had to issue a retraction, and I contacted Quantum and spoke with Eric Bassier, senior director of data center products and solutions, who set the record straight. Seems Quantum -- like IBM and Spectra Logic -- is excited about LTO-8 tape technology and how it can be wed to the company's Scalar tape products and StorNext file system.
Bassier said Quantum was "one of only a few storage companies [in 2016] to demonstrate top-line growth and profitability," and its dedication to tape was not only robust, it succeeded with new customers seeking to scale out capacity. In addition to providing a dense enterprise tape library, the Scalar i6000 has 11,000 or more slots, a dual robot and as many as 24 drives in a single 19-inch rack frame, all managed with web services using representational state transfer, or RESTful API calls.
Quantum, like IBM and Spectra Logic, is articulating a product strategy that has fingers in all the popular storage buckets.
Quantum was also hitting the market with a 3U rack-mountable, scalable library capable of delivering 150 TB uncompressed LTO-7 tape storage or 300 TB uncompressed LTO-8 in storage for backup, archive or additional secondary storage for less frequently used files and objects. Add compression and you more than double these capacity numbers. That, Bassier asserted, was more data than many small and medium-sized companies would generate in a year.
Disk also has a role in Quantum's world; its DXi product provides data deduplication that's a significant improvement over the previous-generation model. It offers performance and density improvements through the application of SSDs and 8 TB HDDs, as well as a reduction in power consumption.
All the storage buckets
Quantum, like IBM and Spectra Logic, is articulating a product strategy that has fingers in all the popular buckets, including tape, disk and NVMe flash storage. After years of burying their story under a rock by providing OEM products to other vendors who branded them as their own, 90% of the company's revenue is now derived from the Quantum brand.
Bottom line: We might eventually get to an all-silicon data center. In the same breath, I could say that we might eventually get that holographic storage the industry has promised since the Kennedy administration. For planning 2018, your time is better spent returning to basics. Instead of going for the shiny new thing, do the hard work of understanding your workload, then architecting the right combination of storage and software to meet your needs. Try as you might, the idea of horizontal storage technology -- one size fits most -- with simple orchestration and administration, remains elusive.
