Overview
Explore a cutting-edge far memory system called Carbink in this 15-minute conference talk from OSDI '22. Dive into the challenges of designing fault-tolerant far memory systems for data centers and discover how Carbink addresses these issues through innovative techniques such as erasure-coding, remote memory compaction, and offloadable parity calculations. Learn about the system's architecture, its application interface using remotable pointers, and the advantages of erasure coding over replication. Examine how Carbink tackles the challenges of varying object sizes and efficient swapping, and understand the role of remote compaction in defragmentation. Compare Carbink's performance to existing solutions, focusing on tail latency improvements and overall application performance gains. Gain insights into the future of memory-intensive applications in data centers and the potential impact of fault-tolerant far memory systems on computational efficiency and storage optimization.
Syllabus
Intro
Memory-Intensive Applications in Data Centers
Application Interface: Remotable Pointers
Replication vs. Erasure Coding
Challenge 1: Remotable Objects Have Different Sizes
Challenge 2: Efficient Swapping under Erasure Codir
Remote Compaction for Defragmentation
Evaluation Overview
Tail Latency (Microbenchmark)
Carbink Summary
Taught by
USENIX