This blog post was authored by Eden Zik.
On January 19th, Vertica engineers joined the North East database community for the North East Database Day conference organized annually at MIT, sponsored by Facebook and Microsoft and featuring Turing award winner Michael Stonebraker. The full conference program can be found here: http://mitdbg.github.io/nedbday/2018/
This year Styliani Pantela from the Catalog and Distributed Engine team at Vertica presented her work on Schema Evolution through Optimistic Concurrency Control in the session on data cleaning and schema evolution, chaired by Professor Alexandra Meliou from University of Massachusetts, Amherst.
All Vertica customers use schema evolution to evolve their database schemas, and in particular the add column operation and they ultimately care about the performance. Vertica’s prior solution had a complicated design and made it hard to implement complex features on top of. It was also incompatible with Vertica on shared storage and exhibited performance degradation due to high lock contentions. Styliani’s solution was to implement optimistic concurrency control, a method applied to transactional systems that assumes that multiple transactions can frequently complete without interfering with each other. The solution yielded 3x performance increases in certain cases. In addition to that, the new design enabled new features and made the current implementation compatible with Vertica’s implementation on shared storage.
Eden Zik from the Cloud Storage team at Vertica presented a poster about his work on metadata durability on shared storage, allowing users to utilize “spot instances” to deploy Vertica on AWS – and save costs on compute while maintaining durability guarantees. This work is part of Vertica’s new cloud offering, known as Eon mode.
Stay tuned for next years conference!
Styliani during her talk on Schema Evolution through Optimistic Concurrency Control
Eden presenting his poster on Eventual Durability for Cloud Databases on Ephemeral Nodes