Over the last few years, much of the enterprise storage market has focused on flash technology, but the hard disk drive (HDD) format has plenty of life in it yet.
It is true that flash storage offers dramatic performance improvements over conventional hard disk drives. And flash memory drives have fallen in price and grown in capacity. Gartner, for example, expects prices to fall between 10% and 15% this year.
Meanwhile, newer flash implementations have helped boost SSD use. This includes NVMe flash, which connects to the host CPU over the PCI Express bus and provides higher read-write speeds with lower power consumption than SATA – or SAS-connected SSDs.
And the ability to use NVMe as networked storage – NVMe-over-Fabrics – has moved flash beyond single server and direct-attached deployments.
All this might suggest that “legacy” storage formats such as hard disk drives and tape have had their time, but this is far from the case.
Still life in spinning disk?
IT departments have more choice than ever. Costs of solid-state storage have fallen to the point where there is little difference between hardware with an SSD or an HDD, especially for endpoint devices. Popular manufacturers such as Dell and Lenovo use flash in almost all their enterprise laptops.
But the economics and even the performance advantages are less clear cut when it comes to enterprise storage subsystems.
Flash memory suppliers have driven down costs and bolstered capacity, but so have hard drive makers. For now, flash media cannot compete on capacity alone.
And careful deployment of storage arrays, especially in SAN and NAS environments, can narrow the speed gap to a point where it is largely irrelevant to enterprise applications. RAID and erasure coding technology ensures that HDDs are also robust.
“If speed is essential, don’t use spinning disks,” says Tony Lock, analyst at Freeform Dynamics. “But you have to be willing to pay for performance. You will want top-of-the-line NVMe or NVMe-over-Fabrics, but there are lots of applications that are happy to run on spinning disks, or on hybrid storage with flash and flash cache.”
NVMe, where storage connects to the host CPU over the PCIe bus, is now established as the “gold standard” for high-performance data storage. Available for storage arrays and internally in servers, NVMe removes the data transfer bottleneck caused by the hard-drive era SATA connection, and maximises the performance benefits of flash.
As an industry standard, NVMe connects to the various types of Nand flash media, so system designers can pick the right mix of performance, price and durability.
And the incorporation of NVMe into storage networks, through NVMe-over-Fabrics, allows all-flash NAS and SAN array products.
SATA-based flash drives remain an option for lower-cost, lower-performance systems and for upgrades to legacy hardware, previously designed for HDDs.
Hard drives, for their part, continue to connect to storage systems through older protocols such as SATA and SAS. Deploying 15,000rpm SAS drives can give throughput of 230Mbps, although the theoretical maximum performance of a four-lane SAS interface is 48Gbps.
Although a SAS drive cannot match NVMe’s raw read-and-write performance, over a NAS or SAN with careful use of RAID levels and flash-based cache, it comes close – or close enough – for a typical enterprise application.
“Hard drives are bigger, cheaper and good enough for most workloads,” says Freeform’s Lock.
Hard drive makers also continue to push capacity upwards. Although there are some limitations to magnetic disk storage (mainly the areal density and the performance of the read/write heads), 16TB drives are now widely available. Seagate has suggested 100TB drives could be possible within a few years, through a technology called heat-assisted magnetic recording (HAMR).
Solid-state media is quickly growing in capacity, but anything approaching HDD levels in raw storage terms is either very expensive or requires a sub-system combining multiple flash drives, or both.
But the main point in favour of hard drives remains their cost. At the top end of the spectrum, Samsung offers a 15.3TB TLC drive, for an eye-watering £3,000 ($4,000). A 16TB Seagate SSD, by comparison, retails for £400 ($500).
Even at the lower capacities used in most enterprise arrays, spinning disks retain a cost advantage. The difference can be anywhere between 4x and 10x per GB, although much depends on the type of flash storage including its durability, performance and interface.
Hard drive use cases
As a result, it is no surprise that enterprises still buy hard disk drives and here are some areas where HDDs still make good IT and business sense:
Flash storage is rarely cost-effective for backup. Increasingly, businesses back up flash media to lower-performance HDDs, on-site for local copies, off-site to a datacentre or DR site, or to the cloud. Today’s backup software compensates for the lower performance of hard drives by running tasks in the background. An HDD-based NAS drive is an economical way to provide local office backups for servers and workstations, as well as staging off-site backups.
Archiving data is increasingly important to enterprises, both for compliance reasons and for future analysis. On-site and off-site archives remain much cheaper on HDDs. And correctly configured HDDs can be as durable as solid state media, which degrades over time through multiple writes.
Analytics and R&D
Developments in analytics are blurring the line between long-term archives and warm and cold data storage.
For analytics and research, the trend to store more data for longer favours the greater capacity of HDDs, at least for now.
The increasing use of sensors and the internet of things by businesses also generates vast volumes of data. For businesses that want to gain maximum use from the data they capture and keep, capacity and cost outweighs raw performance.
Mainstream enterprise applications do not, largely, suffer from storage bottlenecks. Systems such as customer relationship management (CRM) and enterprise resource planning (ERP) run well enough on the current generation of SANs.
Upgrading to faster storage will improve performance, but LAN infrastructure can then become a bottleneck. HDDs work well for application that require sequential access, leaving flash storage best suited for advanced analytics, artificial intelligence, high-performance computing and in-memory databases.
Lower-tier storage to support high-performance systems
Even in demanding environments, only the most frequently used data is stored on flash. The rest is tiered to HDD, cloud or tape. Tiered storage moves data seamlessly between media, with no discernible impact on application performance. Tiering has its own overheads – systems must be well designed – but few high-performance systems use little enough data to work with flash alone.