Add Toshiba America Electronic Components to the list of drive manufacturers working on devices that combine storage,...
compute resources and Ethernet connectivity to scale out object stores, big data analytics and active archives.
At last week's OpenStack Summit in Vancouver, Toshiba demonstrated new drive technology that combines two 2.5-inch high-capacity SATA hard disk drives (HDDs), two high-performance M.2 solid-state drives (SSDs), two Gigabit Ethernet ports, a four-core 64-bit processor and the Linux operating system in a 3.5-inch hard drive enclosure.
Tar Thirumalai, director of market development for Toshiba's storage products business unit, said the new "intelligent" key-value-based drive technologies will provide an alternative to the "inefficient" scale-out object storage model that has "way too many software layers." He said applications write directly to the drives and store additional metadata to speed information access and ultimately cut down on infrastructure costs.
"We want to see how the market reacts to this and then at some point turn this into a product," Thirumalai said. He added, "You can essentially run some of your storage applications directly on this drive, and it would let you store that onto the hard drive as well as onto the SSD."
In addition to the high-performance model, Toshiba also unveiled a capacity-centric Ethernet-based HDD-only version optimized for the shingled magnetic recording (SMR) interface. High-capacity SMR key-value technology is better suited to archival and cold storage applications.
Toshiba isn't the first with this type of technology. At the 2014 OpenStack Summit in Atlanta, Western Digital's HGST division demonstrated its new open Ethernet hard drive architecture, with integrated CPU and memory resources running Linux, to enable applications to run closer to the storage resources than conventional options. HGST said the open Ethernet drive architecture works with HDDs and SSDs.
Months earlier, in October 2013, Seagate Technology introduced its Kinetic Open Storage platform with claims that the technology would enable applications to talk directly to the storage device and eliminate the traditional storage server tier. The company shipped its first near-line Kinetic HDDs in late 2014.
Seagate described its Kinetic drive as a key-value server with dual Ethernet ports that supports the basic put, get and delete semantics of object storage, rather than read-write constructs of block storage. Clients access the drive through the Kinetic API that provides key-value access, third-party object access, and cluster, drive and security management.
Thirumalai said Toshiba used the key-value API that Seagate open sourced rather than "reinvent the wheel." Object storage software such as OpenStack Swift, Ceph, SwiftStack and Scality's Ring already supports the Kinetic API.
He said, with the traditional storage, the system writes data in sectors of a block size, such as 4K or 8K, and the application needs to know the location of the data. With the key-value object technology that major cloud providers use, the system stores a key and a value and does not need to know the physical location of the data to retrieve it, he said.
"The key value makes everything scalable. It's easier to manage. And it also compresses the stack," he said. "You no longer need to have a file system or a volume manager or RAID to talk to storage."
Sage Weil, Ceph principal architect at Red Hat, said via an email that the new devices have the potential to "radically change" the total cost of ownership for large storage clusters by replacing the current "big server with many attached drives" model with "many microserver drives." He said automation tools will become increasingly important because there will be more drives, or "servers," to manage.
"Which use cases these are used for will depend on the workloads and performance requirements," Weil wrote. "The exciting thing about [Toshiba's] KVDrive is that they are going to be quite fast."
Weil said Ceph or Gluster could run directly on Toshiba's KVDrive. He said there would be no server, just drives plugged into disk enclosures with an Ethernet backplane, instead of SATA/SAS, and network switches. He said the drives that HGST demonstrated last year were similar in that they were general-purpose Linux hosts that could run Ceph, Gluster and Swift directly on the disk.
Benjamin Woo, managing director at Neuralytix Inc. said Ceph, or a portion of the storage stack, might run on the new Ethernet hard drive technology, but users won't be able to push everything to it.
"If you start to do too much on the drive level, then you have a problem with network chatter, because all of these drives would have to talk to each other," Woo said.
Toshiba's Thirumalai said the goal is not to displace servers. He said one motivation to work on the new drive technology was to change the way enterprise infrastructure is deployed and bring down the overall total cost of ownership. He said pricing is not available at this point.
"At a base level, we think it would be cost competitive to having a Xeon server with say 18 or 20 drives," Thirumalai said. He said Toshiba plans to work server OEMs and ODMs, but he provided no timeline.
Robin Harris, chief analyst at StorageMojo, said the new drive technologies represent "an extreme form of hyper-convergence," but focused on storage. He said the servers are left to focus on computation, and the powerful disk microcontrollers manage most of the storage overhead.
"Moving intelligence into storage devices has been a 50-year secular trend that got interrupted by RAID controllers," Harris said. "Toshiba's and Seagate's work here is a continuation of that trend using technologies undreamed of 25 years ago."
Ashish Nadkarni, a research director in the storage practice at International Data Corp. in Framingham, Mass., sees the trend is happening because of the emergence of key-value storage devices, such as object storage and NoSQL databases, which can offload key-value computations to the drive.
"It is only being held back by the lack of open standards," Nadkarni wrote in an email. "Once that happens, it will take off."
He said the Storage Networking Industry Association recently started a standardization effort, with Seagate and others supporting it.
Woo said the new Ethernet hard drive technology will enable enterprises to grow and scale more quickly, reduce power, footprint and cooling, and lessen the impact of small components failing.
"It's going to take time for people to understand this technology. For your average IT guy, when you draw the block diagram of how these drives work, it is very confusing," he said. "And I think that Seagate, Toshiba and anybody else jumping on the Kinetic platform have to start to help customers to understand this technology much more easily."
Are enterprise hard drives being pushed to the middle tier?
Experts share predictions on future of flash drives