TECHNOLOGY PAPER
INTRODUCTION
Seagate, as a leader in storage technology, is releasing to market MACH.2!" multi-actuator technology hard disk drives (HDD). HDDs provide the storage industry with consistent capacity growth at a consistently declining cost, making them a dominant storage technology across all markets and applications. Enterprise
3.5-inch, 7200-RPM HDDs, commonly known as nearline drives, have doubled capacity in the last five years, enabling the lowest cost, while maintaining a consistent level of performance. These HDDs have met the growing exabyte demands of cloud infrastructures. As HDDs deliver more capacity at lower costs in the future, to extract maximum value from these storage devices, a consistent performance level must be maintained. MACH.2 multi-actuator technology is on an innovation vector that will maintain performance to match the capacity growth of these 3.5-inch high-capacity HDDs while meeting or exceeding total cost of ownership (TCO) goals.
HDD PERFORMANCE
HDD technology involves moving parts, and performance is driven largely by RPM and the number of channels used to transfer I/Os (read and write commands). Historically, HDDs have gained performance by increasing RPMs, which led to 2.5-inch mission-critical 15K- and 10K-RPM drives. But these 2.5-inch drives cannot deliver higher capacities due to form factor constraints. Similarly, 3.5-inch HDDs deliver the highest capacity but cannot be spun at higher RPMs due to higher power consumption and not enough performance gains to justify higher power. Other challenges at higher RPMs include chassis-level vibration and operational shock performance degradation.
An alternate means of improving 3.5-inch HDD performance is via caching within the drive, which helps only at certain workload conditions and provides limited benefits. Queueing is another means of deriving more performance from a drive, but it comes at a latency penalty and having to architect applications to work at higher queue depths (QD). You may also increase random read/write performance by reducing the maximum usable capacity of a drive (also known as short-stroking) which limits the seek distance across the disk. This technique, though effective for gaining additional performance, is not cost-efficient due to the amount of capacity lost.
1
PROBLEM STATEMENT: IOPS/TB AND TCO
Cloud customers deploy large fleets of 3.5-inch HDDs in their data centers and manage their application storage nodes across multiple devices spread across multiple racks. In order to meet their Service Level
Agreements (SLAs) these cloud customers need to achieve a certain level of performance, measured in
IOPS (I/O per second) and command latency, across the deployed HDD storage capacity. This required performance varies by workload, but can be characterized as IOPS/terabyte (IOPS/TB) at a particular latency.
For many cloud workloads, IOPS/TB is simply derived from the drive's random IOPS capability at a certain workload divided by the available capacity. As long as a particular HDD device meets or exceeds a threshold
IOPS/TB as required by the customer, all of the capacity on the HDD device can be utilized. If the IOPS/TB
on an HDD is below a threshold value as defined by a customer based on application workloads, then the customer cannot utilize all of the drive capacity with the targeted workload. This problem of not being able to utilize all of the capacity due to IOPS limitation at a targeted workload then impacts another critical metric of
TCO: $/TB of utilizable capacity.
14.0
12
12.0
10
