This post is part of a sponsored Pure Storage blog post series. To learn more about Pure Storage, please visit purestorage.com.
We recently covered sustainability and efficiency of enterprise storage architectures in a primer article. We concluded that optimized architectures built on flash media provide more value, both from a performance and overall efficiency perspective.
Once reserved to the most demanding workloads and largest budgets, flash-based storage has democratized and is now prevalent among organizations – and workloads – of all sizes. Nevertheless, it is still common to come across different implementations of flash-based storage architectures.
Flash Storage Architecture Types
We distinguish two architecture types: optimized vs. retrofitted. Retrofitted architectures refer to storage systems that predate the introduction of flash media. They were initially designed to get the best out of hard drives and saw improvements after flash was introduced: first through caching mechanisms, then by adding flash drives to the capacity pool.
In contrast, optimized architectures are flash native solutions that were engineered to take advantage of flash media advantages (throughput and latency) but also mitigate some of its constraints (low durability for some flash media types, for example).
Challenges of Retrofitted Flash Architectures
When building highly optimized architectures, efficiency at all levels is essential. Be it from a performance, resiliency, efficiency, or power consumption perspective, all details matter.
Even within flash storage solutions competing in the same segment, not all architectures are equal. From a vendor perspective, the huge upfront cost of developing and manufacturing highly integrated solutions can be perceived as a financial risk. Even though efficiency is lost, the broad availability of commodity off-the-shelf components promises economies of scale, and cost factors have dictated design decisions. Proprietary storage appliances directly depend on these choices, but software-defined storage solutions are also impacted by their direct dependency on readily available commoditized components.
The adoption of standard (or slightly modified) x86 server chassis fitted with SSD drives (whether SAS or NVMe) has one key architectural implication: each drive is addressed as a standalone entity, with its own controller chip, memory, and Flash Translation Layer (FTL), an area of SSD drives that oversees low-level operations such as wear leveling, bad block management, garbage collection, etc. Media-level operations need to be carried locally on each flash drive, reducing the overall efficiency of the solution: even the fastest NVMe SSD drive is bound by the legacy of HDD architectures. Thus, a major performance bottleneck has been carried over from the old disk-based architectures into the flash world. And as a result, many retrofitted flash architectures require more SSDs of lower density to meet real-world performance requirements, even if larger drive densities are available.
Benefits of Optimized Flash Architectures
Optimized flash architectures are “born for flash”, they do not carry over the legacy of the pre-flash era. Instead, these solutions look at flash efficiencies and constraints, and architect the solution around it to deliver the best possible value.
For example, Pure Storage, a flash-native innovator, understood these challenges and came up with an architecture using proprietary DirectFlash modules and other key architectural innovations that offer a modern, system-level approach to flash storage.
This approach eliminates the old drive-level management paradigm: a single global flash translation layer replaces a multitude of drive-level FTLs, removing the need for non-optimized flash management operations carried independently and blindly on multiple SSD drives (and without touching on other inefficiencies such as on-board DRAM and flash overhead on SSDs to handle FTL needs).
Instead, flash management operations are now global and handle media status and wear holistically. Write operations are optimized to avoid unnecessary P/E (Program/Erase) cycles, increasing the longevity of flash media on one hand, and avoiding the performance degradation induced by P/E activities; written data follows patterns to improve future read operations as well, providing end-to-end performance efficiencies. Other operations, such as garbage collection are now handled globally as well. Pure Storage has an outstanding blog post explaining in detail Pure’s approach to flash. Combined with PurityOS enterprise-grade data reduction mechanisms, the solution offers superior capacity without comprising on performance in a denser form factor when compared to non-optimized solutions from competitors.
Other key architectural differentiators include a tighter integration between components, driving further efficiencies, but also the ability to perform non-disruptive hardware (flash and controller) upgrades, a feature that increases the solution’s longevity and overall energy efficiency, but that will be covered in a dedicated blog post about consumption options and lifecycle management. Finally, it’s important to also highlight footprint optimization: the solution’s components (physical form factor and DirectFlash module layout) are designed to maximize capacity, be it in count of DirectFlash modules, or on the total raw flash capacity that a module can support.
The same principles apply to capacity optimized systems such as FlashArray//C appliances. DirectFlash efficiencies coupled with data reduction capabilities ensure optimal usage of QLC flash media. Organizations can reap the benefits of capacity dense QLC flash without having to worry about media durability, an issue that would otherwise affect solutions based on drive-level flash management.
Even if flash storage is now the standard for more than just performance-oriented workloads, not all solutions are truly flash native. Unconsciously, architectural decisions to go with readily available components bring forth the heavy legacy of retrofitted systems. Performance claims driven by the inherent speed of NVMe SSDs barely masquerade system-level inefficiencies from a capacity, durability, or physical data center footprint perspective, all key contributors to the energy efficiency footprint of a storage solution.
Combined with Purity software and its leading data reduction capabilities, Pure Storage offers a range of storage solutions that provide outstanding capacity and performance in a compact data center footprint, thus enabling the vision of an efficient all-flash datacenter, and delivering best value to organizations from a watt/performance and watt/capacity perspective.