SBIR Phase I: Scaling Operational Intelligence Graphs with GPU Clouds

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) Phase I project will, most directly, be to provide institutions such as banks and national defense agencies with visibility into their graph-­centric security and operations data, and thereby improve their robustness and resilience. Farther out, enabling visualization of larger graphs will also aid non-­IT uses, such as to help financial analysts understand markets, marketing and sales teams understand customers, and precision medicine researchers understand gene interactions. Furthermore, generalizing the underlying GPU cloud infrastructure to scale interactive visualizations beyond graphs will help even more types of analysts comprehend data from an even wider variety of data sources. Likewise, generalizing the proposed GPU cloud infrastructure will also aid non­-visual analytic tasks, such as machine learning over big data.This Small Business Innovation Research Phase I project's goal is to scale visual graph analysis to enterprise level security analytics. Modern graph visualization tools handle at most 50,000 nodes, but many enterprises manage over a million devices and services. The proposed innovation scales graph visualizations to 1­-2 magnitudes more data. The visualizations interact with a GPU cluster for analysis (e.g., cross-filtering), visual layout (e.g., ForceAtlas2 and edge bundling), and rendering. Key to Phase I is establishing the technical feasibility of building and deploying the distributed GPU cloud architecture. The architecture utilizes novel GPU components and optimizations for: (i) GPU analytics (ii) GPU-accelerated visual layouts; (iii) cloud GPU resource management; and (iv) a streaming renderer to draw more while maintaining perceived quality and responsiveness. In collaboration with pilot customers, the views and workflows necessary for minimal viable security analytics usage will be identified and prototyped.