Defining the Software-Defined Technology Market
September 15, 2014, Storage Developer Conference, Santa Clara, CA—Mario Blandini from HGST described the development of a software-defined disk drive that adds CPU and memory to the storage to move compute closer to the storage. This configuration allows for greater flexibility and the ability for apps to run storage directly rather than through a disk drive interface.
The driver for the development is the tremendous growth in data across all areas. In addition, the value of those data is increasing as companies work to mine their complete data set for insights. Active data are becoming longer term as the mining capabilities improve, and readily available data are more valuable than unavailable data. For a typical analytics use case, access time is about 1 second so any improvement in latency or analysis time is significant.
The challenges for the datacenter are scalability, cost, and accessibility, and in the past you get to pick any two. Earlier generations tried for capacity and cost, but not accessibility, so the deep tiers were not easily nor quickly accessible. The advent of active archive has changed many underlying assumptions and is providing high scalability and accessibility with a different cost model.
A combination of optimized HDDs, tuned enclosures, device affinity changes, and scale out software is leading to more fully software-defined systems. The open Ethernet drive architecture is a foundational building block for the software-defined datacenter. The drive architecture is based on a Linux-based CPU plus RAM and Ethernet connectivity to run storage services directly on the storage device and connect directly into the datacenter fabric.
A demonstration drive uses a standard 3.5inch HDD form factor and an ASIC that integrates an ARM CPU and Ethernet with drive electronics. The reference design fits into a 4U enclosure with 60 drive slots and an embedded switch fabric. The chassis allows for hot swap of all components and integrated into existing datacenter architectures. The enclosure has 10Gbps Ethernet and makes the devices look like Linux servers.
A detailed view of the system shows a 32-bit ARM with 512K cache, 2GB DDR3 with 1792 MB available to Linux, a block storage driver to enumerate as a SCSI disk (/dev/sda) and an Ethernet network driver that is enumerated as eth0. The first demonstration units look like any standard 4TB drive.
Various software organizations and companies like Redhat and openstack, have software solutions for this type of smart storage. Open management frameworks like Swift, ceph, and Gluster also support the enhanced drives. Other drive manufacturers have joined in the efforts and have demonstrated interoperability in mixed environments. Some development information is available at www.hgst.com/opendev
