- Big Data Platforms
Big Data Platforms
One of the key design goals of Bigstream is that acceleration must be achieved without changing application code or requiring special APIs. That means that Data Scientists, Developers and BI users can use their standard tools while Bigstream provides the Hyper-acceleration.
Bigstream Hyper-acceleration Layer
- Accelerates Big Data platforms without impacting developer code. Transparent by design: Zero Code Changes
- Native Big Data platform APIs are translated into an internal Dataflow. That means acceleration Across Big Data Platforms
- Advanced Gorilla++ compiler technology is used to Intelligently Route Computations to the best processing platform
- A processing Hypervisor enables performance boost across different hardware (CPU, GPU, and FPGA)
- This results in accelerated performance from 3X – 30X Depending on the processing hardware
Bigstream xRay Console
The Bigstream xRay Console provides a view into the inner organization of your Spark application and the underlying processing microarchitecture.
The xRay console can identify the best Hyper-acceleration opportunities for any given application and can help take maximum advantage of the available processing platforms.
Bigstream Hypervisor is responsible for slicing the computation between different hardware (CPU, FPGA, and GPU), as well as routing data to these hardware resources to ensure the highest performance the data reuses.
Bigstream Dataflow is a platform-agnostic and hardware-agnostic representation of computation for Big Data and Machine Learning applications.
Bigstream software acceleration generates C/C++ code from the dataflow representation of the computation, which is then natively compiled. 2X-5X acceleration is typical.
Bigstream hardware acceleration programs FPGAs and GPUs using dataflow representation of computation, yielding very high performance. 10X to 30X+ acceleration is possible.
Dataflow Adaptation Layer
The Dataflow Adaptation Layer converts native dataflows generated by Big Data platforms like Spark and Kafka into an internal Bigstream dataflow.
- FPGA Processing Platforms
Bigstream Hyper-acceleration Layer makes intelligent decisions on which processors should handle specific computing tasks. This happens transparently and does not require any interaction or configuration by users or administrators.