JumalaSika ltd - Fotolia
The move – announced last week at the flash pioneer’s Pure Accelerate event in San Francisco – will see models in its FlashArray //M range quickly replaced by NVMe-capable FlashArray //X arrays. Until now only the top of the range FlashArray //X70 has supported NVMe storage media.
Pure’s all-flash arrays will now comprise the //X10, //X20, //X50, //x70 models and a new flagship //X90.
Functionality on the new arrays includes NVMe support, connectivity with external storage shelves via NVMe-over-fabrics (NVMf), NVMf over Ethernet and Fibre Channel between servers and storage, support for 10/25/40/50/100G Ethernet and 16 and 32 Gbps Fibre Channel.
Pure characterises its new arrays as shared accelerated storage, with the insistence they should not be considered all-flash. This is terminology derived from Gartner, which coined the phrase to describe storage systems that support NVMe and are capable of being inserted into “rackscale” disaggregated architectures.
It is in fact largely marketing-speak that describes a new generation of block storage arrays; ones that rely on NVMe and NVMe-over-fabrics and that are capable of delivering high levels of performance and low latency and that can act as storage arrays and an alternative to server HDDs.
Several Pure customers, such as ServiceNow and Workday, spoke at the San Francisco event about how the new arrays boosted density, simplified infrastructures and made them more reliable as well while eliminating drives on application servers.
The FlashArray//X family
The new line of FlashArray//X arrays use upgraded x86 controllers – although the firm kept CPU and memory specs under wraps – with the same proprietary NVMe flash modules (DirectFlash) already used in the //X70.
Like its previous generations of arrays, FlashArray //X models will have dual redundant controllers in a 3U chassis.
Pure said that even with the new NVMe drives the new systems won’t be any more expensive that those in the //M range.
That’s good news because most of Pure’s competitors have positioned their NVMe offer as a new top-of-the-range and higher in price than preceding systems.
The entry-level FlashArray//X10 offers raw capacity of 20TB maximum (with 10 flash modules of 2.2TB) for an effective capacity claimed of 55TB. As for the //X20, it can accommodate up to 87TB raw for a maximum capacity of 275TB after compression and data deduplication.
The FlashArray //x50 has maximum raw capacity of 183TB and a theoretical maximum of 650TB. Finally, the new flagship FlashArray //X90 posts a maximum raw capacity of 878TB and 3PB effective.
Read more about NVMe flash storage
- New media provide a range of options to speed workloads, from “old-fashioned” flash to storage class- and persistent memory. We help you exploit the storage performance hierarchy.
- Film giant Technicolor had to duplicate workstreams or schedule access to movies being worked on, but Excelero NVMe allows concurrent access to 10 or more workstations at a time.
The launch of the new arrays is accompanied by the arrival of an external extension shelf that supports NVMe-over-fabrics. Dubbed Direct Flash Shelf, it has a raw maximum capacity of 512TB (or 1.9PB after data reduction).
Theoretically, there’s nothing to stop it being used as a just a bunch of flash accessed directly by servers without passing through FlashArray controllers, but for now, Pure – recognising this method of use is possible – has not decided to sell the shelf independently of its controllers.
Regarding performance, Pure has kept its cards close to its chest about gains expected in the new arrays. No IOPS or throughput specs have been furnished by the array maker for the new systems while specs for existing arrays have also disappeared from Pure’s website.
According to Pure, IOPS specs don’t reflect the performance of the arrays in real conditions and the firm therefore prefers to keep silent about the subject. Pure has indicated, however, that the latency of the new arrays will be close to 250µs and that customers can expect significant gains in throughput.
Coming support for NVMf to host servers will also allow latency to be lowered compared to iSCSI and will simplify deployments.
The deployment of NVMf-over-Ethernet allows, for example, the ability to leverage “any-to-any” connectivity in that protocol and avoid configuration problems related to multi-pathing in iSCSI.
The arrays will come with version 5.1 of the Purity operating system (OS), with the company promising an improvement of around 20% in data reduction capabilities (which will also apply to existing //M systems).