ScaleFlux Briefing Note
Data-driven, processing-intensive workloads like database processing, artificial intelligence (AI) and machine learning (ML), and content delivery networks (CDN) require new levels of storage performance, which today is coming to mean deployment on a non-volatile memory express (NVMe) solid-state drive (SSD). NVMe drives have facilitated so much bandwidth between the flash controller and NAND chips that the Storage Bus (typically in this instance PCIe) becomes flooded. However, application performance still suffers because compute resources have now become a bottleneck.
Potential workarounds include deploying NVMe drives in a direct-attached approach within the server and adding more advanced networking capabilities like Smart Network Interface Cards (NICs). However, both of these approaches add costs and complexities, and they do not address the core issue at hand. As a result, resource utilization remains inefficient.
However, computational storage is emerging to address this bottleneck. Computational storage is an architecture that places compute directly on the storage media devices, so that data may be processed in place. Computational storage minimizes the amount of data that must be moved between storage and the host CPU, and it enables workload processing to be parallelized across a number of drives. As a result, it enables data to be processed more quickly, and more efficiently.
Who is ScaleFlux?
ScaleFlux is a computational storage provider that focuses on addressing the application performance bottlenecks of storage input/output (I/O) latency and low CPU efficiency. Specifically, its approach is to integrate a field-programmable gate array (FPGA) that can process data on the PCIe Flash storage device directly. ScaleFlux claims it has an advantage over competitors in that the customer doesn’t need to modify or even recompile their application to leverage computational storage. Thanks to the FPGA approach, the drive adapts to the application instead of the application adapting to the drive.
Because FPGAs are highly configurable, ScaleFlux can offer both turnkey application-specific solutions, as well as solutions that are tuned to customers’ unique applications with customized software. The FPGAs also introduce greater infrastructure agility, because they can be configured post-deployment to introduce new capabilities into the environment or to support new applications.
ScaleFlux has focused on avoiding a slowdown to storage performance as a result of adding compute to the storage drive. It applies fixed functions within the FPGAs to increase speed and efficiency, resulting in levels of storage latency that ScaleFlux claims are on par with NVMe SSDs. According to ScaleFlux, when this storage performance is complemented by compute acceleration, it enables ten times the application throughput with only half the capacity of traditional flash drives in hybrid transactional-analytical database applications. Furthermore, this performance is highly scalable; as storage I/O throughput increases and more SSDs are added; more compute is also added, resulting in additional net throughput for the application.
ScaleFlux drives are typically sold through integration into systems with hardware OEM partners. It works particularly closely with the Dell Technologies Extreme Scale Infrastructure group, but also ships products through other original equipment manufacturers (OEMs) and value-added resellers (VARs). According to ScaleFlux, its strongest traction is currently among hyperscale and webscale companies and is steadily gaining proof of concept with more traditional enterprises.
Typical use cases include any application that leverages Aerospike, HBase, MySQL, Hadoop or OpenZFS. Customers with applications written in these environments can experience consistent improvements in latency, IOPS improvements, job completion rates and write throughput.
ScaleFlux is unique in the nascent computational storage market because it focuses not just on moving compute, but also on accelerating and improving the efficiency of compute. Its FPGA-driven approach achieves these objectives by facilitating a processing engine that is customizable for the task at hand. This configurability of FPGAs enables faster time-to-market – a key competitive advantage for ScaleFlux customers given the fast-paced nature of today’s business environment. FPGAs have a history of being expensive and complex to work with, but their price points are steadily coming down, and through the ScaleFlux architecture, there are no extra costs on top of the FPGAs because they plug directly into drives. Also, ScaleFlux has know-how in working collaboratively with its customers to ease the process of tuning infrastructure to specific requirements.