Direct Flash: The new generation of flash memory
Flash devices allow companies to provide fast storage on-premises. This can be particularly helpful for hybrid architectures. Processes that require low latency and high bandwidth are hosted locally using flash memory. Less demanding applications – including “cold” data sets – are migrated to the cloud.
Pure Storage has designed a new generation of flash arrays that release the full power of all-flash storage media. Reading time 3 minutes
More isn’t always better
It has taken a long time for SSDs – and this transition is far from over yet – to establish themselves as conventional storage devices in IT architectures. Although developers have managed to increase the density of SSDs many times over, a problem is looming ahead, that has already been affecting traditional HHDs: the capacity increases when speed does not, resulting in a decrease in performance per TB.
In addition, these high capacity SSDs are linked to a single SAS connector. This leads to a bottleneck in data transmission and I/O operations.
There is another issue with SSDs – and this one brings a whole bunch of problems with it: The software operating an SSD is located on a controller chip within the SSD. So first of all, a portion of the SSD’s storage is unavailable and secondly each SSD of an all flash array is controlled autonomously. We end up having two layers of management: One for the entire array and one for individual SSDs. But these layers hardly communicate with each other. Since the latency of an SSD fluctuates heavily with its workload however, communication is essential.
A central brain
Pure Storage has presented an elegant solution to these problems: DirectFlash. DirectFlash hardware consists of flash modules without controllers or software, i. e. 100% capacity utilization. There is no local management; quite the contrary is the case. The entire array is managed globally by the DirectFlash software. This eliminates the unpredictability caused by intransparent micro-management of individual SSD controllers.
Communication with the flash devices is 100% via NVMe. Non-Volatile Memory Express (NVMe) is an interface for accessing non-volatile storage vial Peripheral Component Interconnect Express (PCIe). Connections over NVMe have higher bandwidth and lower latency than those over SAS.
Global Flash management is subdivided into three functional areas:
|1||Adaptive I/O control: Manages all I/O operations based on module utilization. Implements QoS.|
|2||Smart Endurance: Centralized allocation, wear leveling and garbage collection. Ensures greater efficiency and thus a longer service life.|
|3||Predictive Resiliency: Monitoring of the modules down to the flash block level.|
This new technology is used in the FlashArray//X for the first time. Where the design of the FlashArray//X resembles that of the FlashArray//M, the latter uses SSDs. The good news is: in wise foresight, Pure Storage built in NVMe compatibility in the FlashArray//M allowing customers to upgrade their array.