jules - Fotolia
Long known for its flash memory cards used in consumer mobile devices, chipmaker SanDisk Corp. has muscled into enterprise flash during the past decade. The company has developed branded high-capacity solid-state drives, and markets its all-flash InfiniFlash hardware platform through OEM deals with Tegile, Nexenta Systems and IBM.
In October 2015, Western Digital Corp. announced plans to buy SanDisk in a blockbuster cash-and-stock transaction valued at $19 billion. The two behemoths claimed the merger creates a company with nearly $20 billion in combined revenue and a total addressable market for enterprise storage products estimated to reach $76 billion by 2017.
SearchSolidStateStorage recently talked to John Scaramuzzo, SanDisk's senior vice president and general manager of enterprise storage, about the pending merger, high-capacity flash media and emerging flash technologies, such as NVMe over Fabric and ReRAM.
How have you seen the emergence of flash change the way organizations approach enterprise storage?
John Scaramuzzo: The focus on data acceleration during the past five to seven years is the result of fast flash storage. When the first enterprise SSDs were introduced by sTEC and EMC [as a way of] replacing hard drives, everyone in the industry wanted to know when flash prices would cross over on price with hard disk drives. Back then, nobody was talking about cloud storage, because there was no high-capacity flash media to get that much data out of a pipe in the cloud. Flash made cloud storage possible.
The first incarnation of flash was as cache [used with] hard disk storage, which is the paradigm for [on-premises] storage. We are still in the early stages of optimizing applications with flash. As flash technology continues to mature, the cost curve is bending downward to make flash more competitive from a price and density perspective.
Flash is still considered a storage medium for performance, but you talk about high-capacity flash. Where does high-capacity flash fit in the data center?
Scaramuzzo: We look at flash storage versus HDDs in terms of cost per gigabyte, and in terms of cost of infrastructure. And that's exactly the way hyperscale data center customers view flash. They want high-capacity flash drives to reduce the amount of peripheral hardware they have to support. You can get 10 TB densities out of HDDs, but you can't get the data access that you get with flash storage. High-capacity SSDs that give performance at a good cost is the combination customers want, especially in the hyperscale market.
Can flash really replace hard disk drives for high-capacity storage?
John Scaramuzzosenior vice president of enterprise storage, SanDisk
Scaramuzzo: Prices are falling due to the high capacity flash has for scaling. Hard drives have reached a plateau. The industry is getting every last ounce out of perpendicular recording and conventional magnetoresistive heads. HDD capacities are staring down the barrel of heat-assisted recording, shingled magnetic, helium ... every trick in the book. What flash gives is true technology scaling. We've scaled a 2D node down to 15 nanometers, and are following that with the first 3D nodes at 48 layers. Then, we'll go to 64-layer, 96-layer and 128-layer form factors. As an industry, we literally are increasing flash densities on a roadmap that doesn't appear to have an end in sight. Flash scaling over the next five to 10 years will continue to produce higher densities -- increasing the output per wafer -- and that directly results in lower flash prices. It's because flash has a better scaling roadmap than hard drives that we're seeing use cases for the two storage media converge.
How close are we to seeing an all-flash data center becoming common?
Scaramuzzo: An all-flash data center, I think, is a predominant theory, an active theory. For the active storage tiers, I believe the top tier is going to use DRAM plus storage-class memory, and the mainstream storage tier will use high-capacity flash that extends all the way down to the active archive tier. Below that, the cold archive tier will be disk, laser [and] maybe tape, in some combination that's more cost-effective.
The news of the Western Digital acquisition of SanDisk shook up the storage industry. What does this pending merger mean for customers of both companies?
Scaramuzzo: I don't think it's very complicated. We have a shared customer base. If you're a hard drive company, you can only solve so many problems. If you're a flash company, you can solve a different set of problems. As a merged entity, we will be able to solve both sets of customer problems. There is tremendous opportunity in both the enterprise and consumer retail markets with products [geared for] the Internet of Things. So, it made sense for the two companies to come together.
How has NVMe over Fabric standards development influenced the SanDisk roadmap? When will SanDisk introduce branded NMVe-compliant SSDs?
Scaramuzzo: We don't have NVMe products announced yet, but, obviously, it's in our plans. That is where the market is heading. NVMe offers opportunities for a lot of optimization when you build a data center. It fits with our goal of making it easier for hyperscale architectures to use and implement storage. It certainly helps to be able to disaggregate the data from the CPU, and processors and servers, allowing large-scale data centers to manage their processors and their storage as independent tracks. That would be hugely powerful.
Which storage products are closest to market in SanDisk's near-term product portfolio?
Scaramuzzo: We announced a partnership with [Hewlett Packard Enterprise] in October 2015 to develop a brand new storage-class memory technology, called ReRAM. We are supplying our ReRAM resistive memory technology to [HPE] to build products with it. ReRAM has a three-dimensional structure that is similar to 3D NAND, but it's faster than flash and cheaper than DRAM.
When do you expect ReRAM to begin showing up in enterprise storage products?
Scaramuzzo: We have given 2018 as guidance. ReRAM is byte-addressable, so it solves the problems [associated] with block-addressable devices, like an ULLtraDIMM or PCIe device. We think storage-class memory is going to revolutionize storage architectures once it gets productized.
The future of flash storage technology
Flash storage will be sticking around in 2016
Flash developments may spell end for NAND
Is an all-flash data center implementation really necessary?