Data Management

Data management brings together MSRG’s work on cloud-native systems, blockchains and distributed ledgers, consensus, and graph data processing. Across these areas, we study how to make distributed infrastructure more scalable and efficient.

Cloud Native Systems

We study concurrent analytical queries in in-memory OLAP databases. Queries can repeat computations and produce the same intermediate data. Our approach detects these overlaps online and coordinates execution so queries can share work and improve throughput.

Blockchain, Distributed Ledger Technology & Consensus

We develop consensus algorithms, fault-tolerance mechanisms, and consistency models for high-performance, scalable, available distributed applications. Our Hyperledger Fabric work examines transaction failures and uses optimization recommendations and adaptive configuration to improve throughput, latency, and transaction success.

Consistency Theory

We investigate reversible operations and Byzantine fault tolerance for conflict-free replicated data types (CRDTs), developing algorithms that extend their replication capabilities.

Graph Processing

Graph partitioning affects distributed processing performance. We use machine learning to select a partitioner suited to the input graph and its workload.

Current directions

• Shared execution of concurrent analytical queries
• Consensus and distributed-ledger performance
• Reversible and Byzantine-fault-tolerant CRDTs
• Learning-based graph partitioner selection