Want to learn more about NVMe in the data centre? Click here to instantly purchase and download our “NVMe in the Data Centre” report, release 3.1 with vendor reviews.
This week at Pure Accelerate 2019, Pure Storage announced the release and general availability of a new FlashArray platform, FlashArray//C. The “C” stands for capacity and a new family of products that use TLC/QLC flash to deliver a high-capacity, lower-cost implementation of FlashArray.
The FlashArray platform has now moved through seven generations, including the latest FlashArray//X models that support NVMe over Fabrics. This evolution has increased performance and capacity, but from a price perspective, FlashArray//X is still a premium product. Moving to an all-flash data centre is a big commitment. Assuming you can take the cost, the benefits are much more than just performance. A common platform provides greater reliability, consistency and availability that affects all parts of the business.
For more details, listen to this podcast with Patrick Smith, Field CTO for EMEA and former customer, who deployed FlashArray end-to-end in his data centre.
Making the justification for placing all block-based applications onto tier-1 flash isn’t easy. There are many scenarios where secondary data could easily live on a hybrid platform that looks to optimise cost using traditional techniques like tiering. This is exactly where FlashArray//C is being positioned – as a tier 2 platform with a cost profile to challenge existing disk and hybrid arrays.
The //C platform looks pretty much like a standard FlashArray (and in fact, most of the components are the same). The major difference is in media choice. Initially, //C will use TLC media, moving to QLC NAND flash early next year, once the availability of QLC becomes more stable (and pricing decreases over TLC).
Today, //C is offered as a single model, the FlashArray //C60, with three capacity points. There’s entry-level 366TB raw (1.3PB effective after data services optimisation), 878TB raw (3.2PB effective) and 1.39PB raw (5.2PB effective). Performance is 2-4ms, with six 9’s (99.9999%) availability. FlashArray//C supports new DirectFlash modules, with 24TB each of initial capacity and 48TB drives expected next year. There is no word on pricing yet.
It’s worth noting that there aren’t many NVMe drives on the market today that use QLC flash. The Intel D5-P4326 series is an enterprise drive available in U.2 and E1.L form factors with 15.36TB of capacity and planned 30.72TB of capacity later in 2019. Other solutions are typically only consumer-class devices.
QLC devices have greater capacity than (for example) MLC flash, packing in more bits per cell to offer greater density. However, QLC drives have lower endurance and lower latency compared to MLC SSDs, with endurance being highly workload-dependent (for example the D5-P4326 has 5x the endurance with sequential vs random workloads). This means putting QLC into an enterprise-class product requires some additional work. Drives are more likely to be oversubscribed and need additional effort to gain a consistent level of endurance and performance. This is the reason we see //C quoted with 2-4ms latency.
Of course, FlashArray doesn’t use standard SSDs. Pure Storage developed DirectFlash, which pushes the FTL (flash translation layer) into software. At the time DirectFlash was introduced, Pure called the hardware a “software-defined flash module”, which seems to be contradictory. However, moving flash management into software allows the Purity operating system on FlashArray (and FlashBlade) to manage media wear and write amplification, extending the life of TLC and QLC media to a level that is acceptable for the enterprise. This is a feature we’re only now starting to see be developed in the wider industry with the introduction of Open Channel SSDs.
What will FlashArray//C be used for? Pure are suggesting a range of use cases, including VM-tiering (using SPBM), lower-cost disaster recovery, multi-cloud dev/test environments and snapshot consolidation. These are all examples of scenarios where Pure Storage possibly couldn’t have played before, especially with customers that weren’t prepared to place all of their data on FlashArray//X.
Of course, simply offering a tier-2 array isn’t enough to gain customers in a competitive market like all-flash storage. FlashArray//C enables Pure Storage to offer businesses the benefits of lower-cost flash with all of the features and functionality already available for FlashArray//X. This is an important consideration for IT departments looking to reduce overall TCO in storage and IT in general.
We’ve seen Pure take the approach of broadening appeal already, with the introduction of ObjectEngine to provide a storage target for data protection and Cloud Block Store for cloud-based applications.
All-flash storage was never going to exist in the data centre as a single flat tier for all workloads. Traditional tiering (in part) was used to address the cost/performance/value equation of placing data onto the most appropriate media for data over its lifetime.
FlashArray//C allows Pure Storage to address a wider market by incorporating secondary data requirements but in a way that doesn’t need two sets of APIs or knowledge of multiple platforms. This will be invaluable for IT organisations that want simplicity in their operation.
One big question that immediately comes to mind is why Pure Storage decided to build out a completely separate product family. Why not build a “hybrid all-flash” platform? FlashArray//X could have simply been extended with QLC shelves connected by NVMe-oF. From the discussions I’ve had, it’s clear that internal array tiering is not part of the Pure Storage strategy. The company believes that internal tiering solutions result in inconsistent application performance.
There’s definitely an argument to be made here between consistent performance and flexibility. In hybrid systems, there will always be a risk of performance variability because almost all tiering models are reactive in nature. Implementing a single storage tier per array is one way to avoid this problem.
What happens if you want to move data around? Virtualised workloads can be managed with solutions like VMware Storage vMotion and automated with DRS and SPBM. Alternatively, Pure offers Active Cluster, which could be made to enable the seamless migration of data between //X and //C arrays. With sufficient data volume requirements, having multiple platforms is easier to justify, especially if the management tools make the transition between two systems an easy process.
The Architect’s View
FlashArray//C fills a gap for Pure Storage and will undoubtedly be targeted at customers with hybrid storage solutions ready for replacement. At Pure Accelerate, I asked the question about whether //C would cut into the market for existing //X customers. Pure doesn’t think so, as the use cases are intended to be different. I have a feeling that there will inevitably be some workload migration off //X platforms to //C, but not a mass migration. Instead, applications that were probably not justified for //X in the first place will be tiered down.
Increasing the total addressable market is a logical move for Pure Storage at this point in the company’s evolution. The acquisition of Compuverde offers the ability to put file services into the mix too. Cloud Block Store for AWS (which we will cover in a separate post) starts the movement towards hybrid cloud. This is a good move for Pure Storage that provides customers with more choice. It moves us another step towards the all-flash data centre.
Pure Accelerate also held a Tech Field Day Extra session. You can learn more on FlashArray//C in this video from the event.
Disclaimer: Pure Storage paid flights and accommodation for Chris to attend Pure Accelerate 2019. There is no requirement to blog or produce content from the event. Pure has no editorial direction over any content produced.
Copyright (c) 2019 Brookend Ltd. No reproduction in part or whole without permission. Post #4D79.