Project Description
The goal of this project is to understand the different dimensions of protecting data in a declustered system of peer nodes. The fully distributed peer-nodes approach offers the promise of great scalability, but it also means that loss of a single node has a transient impact through the full cluster.
Some of the challenges that have proved interesting include:
- Quantitative modeling of data protection under different declustering structures and different models for node and disk failures
- Design of distributed RAID layouts and protocols
- Data distribution techniques to achieve full balance among heterogeneous nodes
- Design of data placement constraints to provide protection isolation across different subsets of data
- Analysis of dynamic hotspots during normal and recovery-mode operation
People
- David Chambliss
- Ohad Rodeh
- Jim Hafner
