Event Stream Processing

Analysis of real-time data stream becomes a must have, time-to-value differentiator, in many domains including Finance, AdTech, Security, etc. Volumes of events increase due to IoT events, Clickstreams, etc. causing to Event Tsunami. Existing systems fail to keep up, running into throughput and latency issues. Rapanda helps organizations to accelerate stream processing:

  • Improving throughput, e.g. running Yahoo's streaming benchmark at 1 Billion events/sec on single FPGA Node
  • Reducing latencies for lightspeed real-time analytics for latency sensitive applications

Rapanda provides acceleration for Big Data streaming processing and Machine Learning facing coming Event Tsunami.


Rapanda Stream Processor

Rapanda provides inline end-to-end Streaming Pipes running on a FPGA. The data-path runs on the FPGA while control and management are kept by the CPU. FPGAs are available as commodity cloud instances part of today’s major cloud providers (e.g. AWS, Azure, and Alibaba) and are integrated with SmartNICs (e.g. Xilinx Alveo U50, Alveo U25). Utilizing SmartNICs, the end-to-end Streaming Pipe processing doesn’t use the CPU. Rapanda's product can be integrated with existing environments, as well as run in standalone mode. Moreover, it can be used on-prem or utilizing FPGA cloud instances and cloud edge instances. Many stream processing frameworks are available today (e.g. Flink, Storm, Kafka and Spark). Stream processing on FPGA extremely improve throughput and latency.

Rapanda Stream Processor Performance

"Yahoo's Streaming Benchmark" becomes a standard method to evaluate real-time streaming throughput and latency. When evaluating a CPU only solution a cluster of multiple servers is used. Rapanda's Stream Processor FPGA based solution, achieving extremely high throughput and low fixed latency: 1B event/sec on and fixed latency of one micro second on Xilinx Alveo card.

Background vector created by kotkoa - www.freepik.com