Title: The computer science behind modern distributed datastoreAbout: AlgorithmsPresenter: Max Neunhoffer (ArrangoDB)Modern databases are scallable and distributed and run on much different hardware than old relational databases.Key algorithms must be changed:Paxos protocol to elect mast and slaves
Parallel sorting algorithms on new hardware to use L2 cache
LSM-trees replaces B-trees to support both sorting and bulk inserts
HLC: clock sync problems
Distributed ACID transactions problems:
How to make sure that:- all nodes agree that transaction happened? (Atomic)
- create consistent snapshot (Consistency)
- hide ongoing activities (Isolation)
- handle lost nodes (Durability)
MVCC instead of ACIDWith ACID: Arango (almost), Spanner, Crockroach
Without: all others i.e. Elastic, Dynamo, Hbase, Mongo, ...Summary: Scallability is more important than ACIDMy rate: :star::star::star::star::star:Links: Preview