Product and service reviews are conducted independently by our editorial team, but we sometimes make money when you click on links. Learn more.

Pure Storage Unveils FlashBlade Scale-Out AFA And //m10

By - Source: Pure Storage
Tags :

Pure Storage announced its new FlashBlade scale-out appliance for unstructured data and its //m10 flash array, which is an entry-level all-flash array targeted at small-to-medium enterprise applications.

The FlashBlade and //m10 allow Pure to capture more diverse all-flash market segments, which is a key improvement that enables it to stay competitive with EMC, its primary rival. EMC recently broadened its all-flash product portfolio, so the Pure announcement is well timed.


FlashBlade is a scale-out storage system designed for unstructured data. The system comes in a base configuration of a single 4U chassis that can house up to 15 blades per chassis. The 4U chassis can store up to 1.6 PBs of effective capacity, and additional 4U enclosures can scale the system to 10s of PBs. 

Pure indicated it will offer a price lower than $1 per GB of effective capacity, which is aided by data compression. The company also indicated that it is achieving up to a 3-to-1 compression ratio, but it does not employ deduplication with the appliance (largely because unstructured data does not deduplicate well).

The FlashBlade denotes a fundamental shift from Pure's typical design ethos, which employed standard SSDs. Each FlashBlade chassis supports up to 15 blades that can store either 8 TB or 52 TB of data apiece.

The blades feature a custom flash implementation that employs an FPGA to manage a large bank of NAND packages. The FPGA conducts ECC and encryption routines, but the onboard Xeon D processor actually manages the bulk of the flash management duties. The 8 TB blade employs an 6-core Xeon D and one FPGA, whereas the 52 TB blade leverages an 8-core Xeon D and three FPGAs.

The Xeon D processors manage the flash through a proprietary protocol over the PCIe bus, and Pure representatives indicated that the system provides more concurrency than NVMe. This is a bold claim; NVMe supports up to 64,000 I/O queues, and each queue supports 64,000 commands. The Pure implementation discards the typical flash virtualization layer, which allows it to map directly to the NAND pages, thus reducing latency. 

The blade also employs a super-capacitor backed NV-RAM device to absorb write traffic during ingest, which has the effect of boosting both the performance and endurance of the underlying flash. The module communicates with the host via the elastic fabric connector on the rear of the blade. This 2 x 10 Gbe connection provides redundancy and multipath capabilities when it connects to a dedicated 40 Gbe switch inside the 4U chassis.

The chassis employs a total of eight internal 40 Gbe switches as part of Pure's Elastic Fabric architecture, which provides custom QoS and routing features. The switches provide blade-to-blade, chassis interconnects and client connectivity with a proprietary Pure protocol that runs over the standard Ethernet connection. 

The internally-developed Elasticity Scale-Out software powers the scale-out object storage implementation, which is an object store that provides compression, encryption and N+2 erasure coding services. The software will support NFS and S3 at launch, with HDFS and CIFS to follow, though the company provided no firm timeline for the latter.

The software creates one namespace with immense scalability; for instance, Pure indicated that a user could add 100 million unique objects or files every second for the next 20 years before reaching its limits.

The FlashBlade allows for seamless linear scaling. Each blade provides the extra compute, performance, capacity and networking required to scale out. The system is available now as an Early Access Program, and a release for production workloads is slated for the second half of 2016.

//m10 and FlashStack Mini

The //m10 is a smaller entry-level version of Pure's flagship //m all-flash array. The //m10 is geared specifically for small to mid-size enterprises and is available for as low as $50,000, which includes a year of support. Users can upgrade the //m10 to the more full-featured //m20, //m50 or //m70 FlashArrays as required. Pure designed the //m10 to provide enterprise-class RAS features for the SME space, such as 99.999 percent availability and no loss of performance during maintenance windows or failures. The //m10 provides up to 25 TB of effective capacity (5 to 10 TB of RAW capacity). Pure Storage allows users to manage the appliance via numerous portals, including the Pure1 mobile app for Android and iOS. The //m10 features an average of <1ms of latency and up to 100,000 32K IOPS.

Pure also introduced its new FlashStack Mini, a converged infrastructure appliance that pre-integrates the //m10 with Cisco UCS servers and networking, and employs either Hyper-V or VMware software. Details are slight on the new converged offerings, but the company indicated that it could support up to 1,200 virtual desktops.

The company also extends its Evergreen support model to the //m10 and FlashStack Mini products, which come to market in June 2016.

Wrapping Up

Pure Storage is known as the scrappy startup that had the chutzpah to take on the established OEMs in the all-flash segment, and it has achieved a great amount of success in its ventures. The move to its own custom flash designs will afford more design flexibility, which will increase capabilities and ultimately reduce cost. The emergence of a small-scale converged offering may also foreshadow future announcements, and is indicative of the current line of thinking at Pure.

The company recently made its successful transition to a public company via its IPO, and its Q4 revenue weighed in at $150.2 million, an increase over last year's $65.9 million. Its losses also declined YoY, from $47.6 to $44.3 million, which indicates solid post-IPO execution. The new enhanced lineup indicates where Pure has been spending its R&D dollars, and it appears to be money well spent. The addition of several new market segments should be another healthy source of growth for the company in the future.

Paul Alcorn is a Contributing Editor for Tom's IT Pro, covering Storage. Follow him on Twitter and Google+.