HDD, Hybrid, All-flash Array or Everything Array

There was a time when the only media you could get in a shared storage array was the humble hard disk drive. Although early systems used DRAM (system memory) as a cache, overall performance was always limited by the capability of the drives themselves, whether that was from the perspective of latency or throughput. Over time, arrays became more diversified with the implementation of tiered storage. Tiering was done to save cost, and it placed drives (and eventually flash) of different performance characteristics into the same array, moving data around using intelligent placement algorithms to determine the optimum location for each piece of data. The aim was to place data on the most appropriate tier to match the I/O requirements of the application, without spending money on expensive storage when it wasn’t needed.

The problem with traditional tiering and dynamic placement was that those systems were always playing catch-up with the application. Data was moved after the fact, once the storage system had measured hours or days worth of performance data. Although these systems were more useful than straight HDD-based systems, they weren’t able to fully exploit the performance of the storage media at the right time.

Fast-forward to today and we have a market with HDD-based systems, hybrids and all-flash appliances. The hybrid devices combine disk and flash in a way that accelerates the performance of the system for only a small increment in cost. The design of hybrid systems allows flash to be targeted more effectively at the active data, either by using it as an extended cache (that can be larger than DRAM was) or through using it to act as the target for I/O journaling. The journaling process means I/O writes can be consolidated to disk as a secondary (asynchronous) process that doesn’t directly impact write I/O. This provides much better write latency than with traditional HDD systems.

The downside to hybrid systems is that occasionally some data may need to be read from disk and so for those requests, performance will be lower than data stored in flash.

At the top of the tree in terms of performance are all-flash systems. These only use flash media and offer the lowest latency and highest throughput for all workloads types. Initially flash systems were very much the “dragsters” of the storage industry, offering little in the way of features and being initially targeted at solving performance issues that would otherwise require application rewrites.

 

Balancing Requirements

Looking at the three platform types (HDD-based, hybrid and all-flash) we have a set of products that meet three basic requirements: cost, performance and capacity.

HDD-based systems are great on solving high-capacity requirements, with today’s HDDs scaling up to 10TB each. These solutions don’t offer the best performance, but are competitive on cost. In practical terms, purely HDD-based systems don’t really exist any more as almost all vendors supplement the hardware with some form of flash to manage performance.

In the middle ground, hybrid systems provide a good balance between all three metrics. More importantly, the better solutions on the market allow the amount of flash installed to be varied, to meet performance requirements. This means performance and cost can be traded off against each other, while maintaining sufficient storage capacity. If performance starts to drop, more flash can be added to resolve the problem.

At the top end, all-flash systems provide the highest levels of performance with an associated cost to match. These solutions typically don’t scale anywhere near as far as HDD and hybrid systems in terms of capacity, as the underlying media is currently only reaching the 1TB-2TB capacities. An all-flash system can deliver consistent low latency for all data compared to a hybrid solution and is therefore good for scenarios where the application latency needs to be guaranteed.

 

Market Evolution

The market for flash and hard drives is constantly evolving. HDDs continue to increase in capacity, although performance remains static. This means the price/capacity ratio continues to improve. Flash devices have moved from relatively expensive SLC media, to MLC and now TLC. At the same time, flash has seen the introduction of 3D in place of planar NAND. This technique places layers of cells onto the silicon substrate, significantly increasing density (and therefore capacity).

The evolving NAND market means there are a range of products to suit requirements with high endurance flash for high intensity write-biased workloads through to “capacity” flash with lower endurance and a correspondingly lower $/GB cost. This means in the future we are likely to see hybrid all-flash systems that use multiple media types to optimize cost, capacity and performance requirements.

Rather than moving to an all-flash world, we are heading towards a scenario where the media used in storage systems could be a mix of HDD, flash or future new technologies such as 3D-Xpoint. The dominant model has to be hybrid because this will continue to allow the mixing and matching of media to suit the three metrics of storage; cost, performance and capacity.

 

Share This Post

More To Explore