Students: Hank Lin, Saurav Singhi, Gourab Mitra, Darshana Balakrishnan

Recently, a swath of specialized data management systems has attempted to displace traditional relational databsaes, each sacrificing a measure of physical independence for the consequent performance gains. However, relying on an entire data management system built around a specific set of performance/capability tradeoffs requires making strong assumptions about (often unpredictable) workload expectations. ASTral does for specialized databse systems what self-describing data did for specialized schemas.  ASTral involves several sub-projects:

Just-In-Time Data Structures

ASTral is based on an idea called Just-in-time datastructures, where data structure manipulation and access logic are decoupled from the physical representation. A just-in-time data structure uses a set of simple semantic and structural building blocks both to emulate the behavior of existing data structures, and to dynamically create new data structures synthesized on-the-spot to match presented workloads.

(The ASTral project is being developed in collaboration with Luke Ziarek, and is supported by NSF Grant #1617586)