Scale-out Edge Storage Systems with Embedded Storage Nodes to Get Better Availability
summary
Claim: replacing server-based racks with many cheap embedded storage devices (networked computational SSDs) yields higher availability because cost/space/power let operators deploy far more independent failure domains. Evaluation uses three-way random replication and an analytical model measuring a relative benefit metric; results report server-based designs have about 7–20x higher data-loss risk than device-based ones in examples, and MBWU-based accounting shows embedded nodes save space and power while raising availability, given the paper’s homogeneity and independence assumptions.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:54Embedded Storage
- 3:23Assumptions
- 4:33Threeway replication
- 5:28System of modal parameters
- 7:16Defining the analytical model
- 8:26Evaluation
- 10:35Summary
- 12:21Conclusion
related talks
44mAvailability, Latency and Cost: Withstanding Outages
35mSystems @Scale Summer Q&A Week 3
47mA Distributed, Scalable and Highly Reliable Data Storage System
20mFine Grained Replicated State Machines for a Cluster Storage System
01mEdge Computing
11mStripeless Data Placement for Erasure-Coded In-Memory Storage
02mStarcluster
42mPostgreSQL Distributed: Architectures & Best practices