Beckman abadi-5min-pres
Download as PPTX, PDF2 likes1,573 views
The document discusses the impact of big data on the database community, emphasizing the need for scalable database systems capable of handling large-scale data analysis and requests. While traditional database technologies are being integrated into platforms like Hadoop, the document warns that the NoSQL movement lacks traditional database capabilities and has not made significant progress in developing distributed databases that can manage transactions at scale. The author questions the feasibility of achieving ACID compliance with active replication in distributed systems, highlighting a persistent challenge in this field.
Beckman abadi-5min-pres
- 2. * The Big Data phenomenon is the best thing that could have happened to the database community * Despite other definitions related to ‘3 Vs’ --Big Data means BIG Data * Which means we need scalable database systems * Still two main components of Big Data * Performing data analysis at scale * Performing requests on data at scale *
- 3. * Database community has won the battle * Some thought that MapReduce might replace traditional database technology as the primary means to perform analysis at scale * Just about every MapReduce vendor has abandoned this goal * Hadapt, Impala, Tez, and several others are in a race to see who can add the most traditional database execution technology to Hadoop fastest * Everyone is going in the direction of cost-based optimizers, traditional database operators, and push-based query execution *
- 4. * The database community is losing the battle * NoSQL systems still have very little traditional database technology inside (despite adding SQL interfaces) * No race to add DB technology --- why? * Don’t blame CAP --- CAP is only relevant when there’s a * network partition We never figured out how to do ACID and active replication at scale * Many new proposals make simplifying assumptions in order to handle scale * It’s been 30 years ---- why can’t we build a distributed database that can handle distributed transactions over actively replicated data at scale? *