Maxim_Kazmin - Fotolia
For all the talk of enterprise flash and the all-flash datacentre, there are several areas where hard disks are still chugging along nicely.
One of those is NAS. You might justify spending anything upwards of £20,000 on terabytes of flash in an all-flash array to power your mission-critical and performance-hungry business apps – but it is rather harder to justify spending millions on petabytes of flash in scale-out NAS, in which data just sits there much of the time.
That's not to say there is no place for flash in NAS systems; quite the opposite. While for primary storage flash is no longer the disruptor – it is the new normal – the question is how much flash is enough for NAS – and how much can be cost-justified?
The same question was asked in the early days of hybrid flash arrays – only this time the likelihood of increasing amounts of flash will proceed more slowly. In part that is because many NAS use cases emphasise the type of storage requirements where spinning disk still wins out, such as deep storage, sequential access and so on.
It also helps that, while in recent years the per-gigabyte cost of flash has fallen faster than the cost of hard disk, the differential may be turning around now, with the arrival of 10TB drives and the like. This is thanks to cost reductions and density increases from new spinning disk technologies such as shingled magnetic recording (SMR) and soon heat-assisted magnetic recording (HAMR) too.
Where flash (and other solid-state storage technologies, of course) can contribute is at the access layer, accelerating file access and therefore the performance of the application by providing advanced caching techniques, such as write re-ordering, metadata storage and other services. Access speed is key in a world where you typically rely on advanced parallel file systems to enable you to store unstructured data volumes that could scale to billions of objects.
Also important is the ability to support tiered storage, with low- and high-performance tiers, and with flash having now pretty much wiped out 15,000rpm hard disks as the primary candidate for the latter. As a result, some specialists report the flash component in NAS systems is up from nothing two years ago to 5% of the capacity they sell today – although they add that the cost per GB differential means this translates to perhaps 10% or 15% in revenue terms.
The amount of flash for the task
But is this enough? That depends on who you talk to. Some argue that you can get big wins with less than 5% of the system's capacity being flash, or a few 10s of GB. That's if you focus it on key tasks, such as storing metadata and small files – and especially in applications where it is not used as primary storage. Others suggest that 10% or 15% – by capacity – would be entirely appropriate, even for virtualised environments with their highly randomised workloads.
However, there are also use cases where all-flash NAS can readily be justified – in particular where NAS is used as primary storage. Alongside performance-hungry applications – such as databases and business analytics – examples might include applications with quality-of-service requirements, such as video-on-demand where 1,000 users might want to stream the same content; or video editing, where you are carrying out I/O-intensive work in very large files.
Some suppliers have had considerable success in these use cases with clustered NAS, where the task of adding flash is eased by the ability to add an all-flash NAS device to the pool, with its solid-state storage then available to all its disk-based fellows. This approach can reduce the cost of replicating to a disaster recovery site, where the all-flash filers can be replaced by cheaper hybrid units.
In the longer term, the bigger threat to disk-based NAS may come from object storage allied to content archiving. In this scenario, TCO reduction comes from aggressive tiering, moving static content into an object repository – where it also no longer needs to be replicated or backed-up daily – and moving the application-centric hot data onto all-flash NAS.
Examples of flash in scale-out NAS
Sonas can intermix solid-state, SAS and SATA disks in its 12-drive storage nodes. These then connect in pairs to interface nodes via Infiniband, and the interface nodes provide network connectivity. With up to 60 storage nodes, 30 interface nodes, a management node and a 96-port Infiniband switch, that is up to 7,200 drives per system. Sonas supports 2.5-inch format flash drives, and can also act as a NAS head in front of other storage, such as IBM's Storwize systems
NetApp's introduction of all-flash filers helped bring the concept of flash-based NAS to a wide audience. The clustering technology in its flash-aware Data OnTap operating system (OS) software allows an all-flash FAS8000's storage to be pooled and shared with disk-based or hybrid filers. Alternatively, up to 24 all-flash filers can be clustered to provide almost 5PB of solid-state storage and millions of IOPS.
EMC Isilon's S-series and X-series NAS devices can use flash drives to speed up file access by storing metadata, to enable in-line data deduplication and to store latency-sensitive data. SSDs can also be used for encryption of data at rest and for data retention. A single NAS cluster can support more than 700TB of globally-coherent flash cache.
Read more about flash storage
- In this special report we look at the product offerings and main trends in the SSD all-flash array market from the big six storage suppliers, as well as the pioneering startups.
- When it comes to choosing between hybrid flash and all-flash storage, the question is increasingly not how much flash is enough – but whether you still need any disk at all.
Instead of the usual 24 SAS hard disks, 2U Compellent arrays can be configured with a mix of 400GB SLC flash drives for write-intensive work and 1.6TB MLC for read-intensive data. Both all-flash and hybrid flash/hard disk configurations are possible. The array's software automates tiering of data between the different storage types.
The object-based StoreAll 8800 appliance and StoreAll 8200 gateway devices support all kinds of flash and spinning disk storage. StoreAll clusters can include different models and even product generations, and scale to 16PB per cluster, with the storage itself coming from HP 3PAR StoreServ arrays. These can support all-flash configurations that comprise hundreds of SSDs each.
Hitachi Data Systems HNAS
Based on technology Hitachi Data Systems (HDS) acquired with scale-out NAS pioneer BlueArc, HNAS operates as a NAS head or gateway in front of storage such as HDS's VSP (Virtual Storage Platform), which includes automated tiering, external storage virtualisation, and flash-optimised system software. Each HNAS node or cluster can scale up to 32PB of storage, and VSPs can be disk, hybrid or all-flash.
Another scale-out NAS pioneer, Panasas offers a range of blade-based hybrid appliances able to scale to more than 12PB. It uses flash for metadata and for small files; anything below 64k is stored on flash while working data is held in an even faster tier of RAM. This enables Panasas to wring significant performance gains from a relatively small flash tier.
DDN ExaScaler and GridScaler
DDN's parallel file systems use solid-state memory for metadata storage and intelligent caching – for example, to re-order random I/O to sequential before it reaches the disk tiers, so that the overall workload performs better. They support more than 20 SSD types, with the choice depending on the characteristics of the workload, the size of its datasets and so on.