Overview
Save Big on Coursera Plus. 7,000+ courses at $160 off. Limited Time Only!
Explore a conference talk on Zero-Change Object Transmission (ZCOT), a novel approach to enhance distributed big data analytics performance. Delve into the challenges of data exchange in high-level languages like Java and Scala, and discover how ZCOT enables direct object copying between Java Virtual Machines without transformations. Learn about the benefits of ZCOT, including minimal integration efforts with existing applications and potential for deduplication. Examine the experimental results showcasing ZCOT's ability to significantly improve data exchange efficiency and boost application performance by up to 23.6% in state-of-the-art data analytics frameworks. Gain insights into the workflow of ZCOT, space construction techniques, compatibility with applications, distributed memory management, and RPC interfaces.
Syllabus
Intro
Distributed Big-data Analytics
Workflow of Big-data Processing
Costly Inter-JVM Communications
Existing S/D Optimizations
How to Achieve This?
Challenges for ZCOT
Space Construction: DCDS
Compatibility with Applications
Distributed Memory Management
RPC Interfaces
The Workflow of ZCOT
Experimental Setup
Mircobenchmark
Flink Performance
Conclusion
Taught by
USENIX