Harvard University's Stratos Idreos is using what he calls a periodic table of data structures to map those structures and their characteristics and to probe their many unique combinations.
He says data system design is shaped by data structures, workload, and hardware, and his team has produced analytical models supplemented by benchmarks and machine learning that can estimate a design's suitability for certain workloads and hardware configurations.
Unlike previous innovations, Idreos' work seeks to not only alter existing data structures, but also reshape them to create new ones dynamically.
He says his team at Harvard's Data Systems Laboratory has worked with RocksDB, reconfiguring the software to realize performance that may be up to 10,000 times better for the same workloads.
The research will be applied to using open source data formats in the cloud.
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA