Abstract:SpeedStream is a universal distributed platform that can handle with massive data flows with the features of low coupling, high availability, low latency and high scalability. Focusing on the core technologies of real-time stream computing platform in cloud environment, this paper conducts a series of researches and implementation of the system. First of all, aiming at the availability of real-time streaming computing platform, we design a high availability framework based on Zookeeper. It ensures fault detection and recovery of process level and node level timely by monitoring heartbreak of each modules and strategy of fault migration. Secondly, in order to increase the application types of the platform, by means of directed cycle detection and iteration protection, we design a real-time streaming computing model that based on directed graph with sources and sinks, which can not only satisfy the needs of common DAG computing services, but also support iteration computing services including directed cycle, bidirectional arcs and annular arcs. In addition, the platform can realize personalized task scheduling strategy for users by establishing task allocation matrix and optimize task allocation model. Finally, in order to solve the many-to-many dynamic load-balancing between tasks, we apply scheduler with status and distributed session table. It overcomes the difficulty of maintaining consistency of session without global session table. We also testified the convergence of this method. The experiment indicates that the throughput and data processing delay of SpeedStream are superior to other alternatives in dealing with the businesses of iteration applications, high traffic fluctuation applications, and high demand of load-balancing applications. This platform provides reliable, universal, and real-time solutions to process massive data flows, such as to process the real-time trading data in e-commerce, to analyze sensing flow in internet of things, and monitor traffics of the Internet.

Streaming supercomputing needs workflow-enabled programming-in-the-large

Scalable-Grain Pipeline Parallelization Method For Multi-Core Systems

Computing Just What You Need: Online Data Analysis and Reduction at Extreme Scales

Parallelizing stateful operators in a distributed stream processing system: how, should you and how much?

Toward Common Components for Open Workflow Systems

Progressive online aggregation in a distributed stream system

Workflows to driving high-performance interactive supercomputing for urgent decision making

Computational Steering and Online Visualization of Scientific Applications on Large-Scale HPC Systems within e-Science Infrastructures

Concurrent and Storage-Aware Data Streaming for Data Processing Workflows in Grid Environments

Streaming Task Graph Scheduling for Dataflow Architectures

A Framework for Stream Programming on DSP

Scheduling On-demand Data Streaming for Cyberinfrastructure Applications with Constraints of Storage and Bandwidth

Bigflow: A General Optimization Layer for Distributed Computing Frameworks

The future of scientific workflows

Streaming Data in HPC Workflows Using ADIOS

Pilot-Streaming: A Stream Processing Framework for High-Performance Computing

SpeedStream: A real-time stream data processing platform in the cloud.

Block-Based Concurrent and Storage-Aware Data Streaming for Grid Applications with Lots of Small Files

Linking Scientific Instruments and HPC: Patterns, Technologies, Experiences

An Efficient Scheduling Algorithm for Stream Computing.

Atomic Streaming: A Framework of On-Chip Data Supply System for Task-Parallel Mpsocs