The much-discussed evolution of computational storage is an exciting transition in the way large data sets are processed. Traditional compute server architectures have been failing to scale with the demands of data users. Today, computational storage is important to all enterprise data managers interested in bypassing performance bottlenecks and maximizing the value of their data.
Fast and Fluid
The real challenge for data managers is to make data access and data processing as fast and fluid as possible. But working with large data sets has historically meant compromise: either processing a small subset of the data quickly, or facing long waits for more comprehensive results. Computational storage resolves this dilemma by pushing intelligence to where the data resides or, in other words, giving the storage endpoint its own processing capability.
Accelerating IO-Intensive Workloads
After years of development, we introduced our computational storage technology, Samsung SmartSSD® drives, designed to improve workload performance and system efficiency. A SmartSSD® drive combines Samsung solid state storage and a flexible Xilinx FPGA compute engine in a single device. There, an internal data path between the storage and the FPGA allows data to be transferred much more efficiently for local processing. This local data processing can involve analytics functions like parsing and filtering, media operations like video transcoding, and standard storage functions like compression and decompression. In each case, local processing is faster, and frees up both host memory bandwidth and I/O interface bandwidth (PCIe lanes) ― two valuable resources that are required for overall processing of high-volume data in large-scale systems. Therefore, with SmartSSD® drives, performance on I/O-intensive workloads can scale to match the volume of data, without bottlenecks.
Put another way: Samsung SmartSSD® drives represent an ultra-fast, energy-efficient distribution of the compute function and a means to smoothly accelerate beyond server CPU limits. Most applications can benefit from being accelerated in or close to the actual storage component: database analytics, video processing, machine learning inference, storage and virtualization, across multiple verticals like media, healthcare and financial services. New market-specific acceleration solutions can be developed easily using Xilinx’s SDAccel environment or by tapping into the ecosystems of SmartSSD® solution partners.
Head of the class
SmartSSD® drives enable faster time-to-insight and excellent scalability. Other gains are reductions in TCO, resulting from lower overall power consumption and the need for fewer CPU cores per system. This should be enough to satisfy even the most change-adverse data manager.
There’s no need to wait to embrace the advantages of computational storage. Contact us here and let us show you just how much it can improve your compute and storage processes. We also encourage you to hear more about SmartSSDs at the 2020 Open Compute Project Global Summit March 4-5 at the San Jose (CA) Convention Center; Booth # B25.