2017年-Talkingdata_【T112017-数据工程和技术分会场】物联网和人工智能领域内置芯片分析的意外之旅_53页_20mb

Future-proofing BI: Leveraging In-Chip Analytics in IoT and AI

Core Content

This document explores the concept of In-Chip analytics as a revolutionary approach to overcoming the traditional memory bottlenecks in Business Intelligence (BI) systems, particularly in the context of IoT and AI. It outlines how Sisense's In-Chip technology enables faster query execution and more efficient data handling by utilizing the CPU's cache memory, rather than relying solely on RAM. The document also discusses the future of BI with a focus on contextual, intelligent, and immersive data consumption, and how it can be integrated into everyday environments.

Main Points

• In-Chip Analytics: Sisense's proprietary In-Chip technology allows for faster query execution by leveraging the CPU's cache memory, avoiding the need to load entire datasets into RAM.
• Memory Bottlenecks: Traditional BI systems face performance issues due to the latency of fetching data from disk to RAM. In-Chip technology mitigates this by using a combination of:
  • Vectorization: Rewriting queries to process data in parallel using SIMD (Single Instruction, Multiple Data).
  • Pre-loading Sub-query Results: Storing and pre-loading compressed results in the CPU's L1 cache.
  • Optimized Data Model: Only relevant columns are loaded into RAM, reducing memory usage and improving speed.
• Performance Gains: In-Chip analytics outperforms in-memory processing, even when handling large datasets. For instance, analyzing 10TB of data can be done in under 10 seconds on a single node of a standard Dell Server.
• BI Everywhere: The document envisions a future where BI tools are not confined to screens but are integrated into physical devices, such as IoT bulbs, for real-time data consumption. These devices provide immediate and intuitive feedback, such as color or light changes, to help users understand KPIs without needing to interact with a dashboard.
• User Behavior Insights: According to surveys, over half of business professionals find visual alerts more effective in driving action. This highlights the importance of intuitive and context-aware data presentation.
• Scalability and Agility: In-Chip enables agile big data analytics by allowing quick and efficient query execution, even with complex analytical tasks such as aggregations, groupings, and top rankings.
• Future Trends: The document predicts a shift toward personal, intelligent, and contextual data interaction, where users can engage with data through voice-activated assistants, augmented reality, and team collaboration tools.

Key Information

In-Chip Technology Overview

• Efficiency: In-Chip recognizes CPU specifications and organizes query data in a way that optimizes cache usage.
• Speed: By storing and reusing data in CPU caches, In-Chip reduces the latency of data retrieval and processing.
• Scalability: It supports large datasets (up to 120M rows, 28GB) without requiring the entire dataset to be in memory.

Performance Benchmarks

• Test 1 (No Concurrency): Analyzing 10TB of data in 10 seconds.
• Test 2 (Concurrency = 2): Maintains fast performance with concurrent users.
• Test 3 (Concurrency = Max): Demonstrates scalability and efficiency under high load.

BI Everywhere Vision

• IoT Integration: Devices like Sisense-Enabled IoT bulbs provide real-time, contextual data feedback.
• Multisensory Data Consumption: Insights are delivered through light, sound, and color, making data more accessible and actionable.
• User Experience: Tools like voice-activated assistants and AR/VR interfaces are expected to become standard for data interaction.

User Insights

• Color-Coding: 57% of business professionals use or would use color-coding for data visualization.
• Alerts: Visual alerts are the most effective in prompting action.
• Future Preferences: 56% of users prefer voice activation/virtual assistants for data consumption.

Conclusion

In-Chip analytics represents a significant leap forward in BI, offering faster and more efficient data processing by utilizing CPU cache memory. This approach not only enhances performance but also redefines how users interact with data, making it more intuitive and accessible through IoT and AI technologies. The future of BI is not just about speed and scalability but also about contextual, real-time, and immersive data experiences that align with modern user expectations.