BIGSTREAM Hyperacceleration for Apache Spark for Samsung SmartSSD CSDs powered by Xilinx FPGAs SOLUTION BRIEF
■Organizations look to Apache Spark for essential workloads in their data environment. As they attempt to scale, though, they typically face challenges of performance and cost. Hardware acceleration represents an opportunity to solve these challenges but introduces the challenge of intricate programming and altering Spark code.
■Bigstream Hyperacceleration solves these problems, introducing software to utilize specialized hardware without requiring any Spark code change.
■The Samsung SmartSSD Computational Storage Device (CSD) represents a shift in the compute-storage paradigm, processing data on the storage device, drastically reducing the volume of data that needs to move to the CPU.
●SCALING WITH DATA VOLUME: A single server can contain multiple SmartSSD CSDs (up to 24 U.2 SmartSSDsin a typical 2U server). Each SmartSSD CSD can run query acceleration in parallel, producing almost linear speedup, even on a low-end or highly over-subscribed host CPU. While NVMe SSDs are fast, most enterprise SSDs are limited to only four lanes of PCI-Express by the U.2 drive connector, creating a bottleneck. By processing data directly on the storage layer, SmartSSD can perform operations using wider interfaces than the U.2 connector supports, or even access storage via a newer PCI-Express generation than supported by the host CPU. By comparison, unaccelerated SSDs stop delivering additional throughput when the host CPU runs out of PCI-Express lanes.
|
|
|
|
Solutions |
|
|
|
Please see the document for details |
|
|
|
|
|
|
|
English Chinese Chinese and English Japanese |
|
2020/11/11 |
|
|
|
|
|
365 KB |
- +1 Like
- Add to Favorites
Recommend
All reproduced articles on this site are for the purpose of conveying more information and clearly indicate the source. If media or individuals who do not want to be reproduced can contact us, which will be deleted.