John O'Brien & Mark Milani | Hadoop Summit 2014
summary
Independent benchmarks run across three clusters compare Hive (0.11/0.12/1.3), Presto 0.57, Impala and InfiniDB on Hadoop 1.3, Hadoop 2.0 and a Cloudera beta, and find that data file format dominates performance — Parquet, ORC, RC or idb and compression often flip which engine is fastest. Evaluation uses three criteria: speed, SQL feature set and cluster-wide scalability. Actian ports Vector to HDFS, uses YARN and patented positional delta trees, promising integrated Vector+DataFlow+Matrix and advising YARN compatibility required for future engines.
AI-written summary · about
chapters (14)
- 0:00Introduction to Hadoop Summit
- 0:28Guest introduction and SQL on Hadoop
- 1:09Radiant Advisors overview and research
- 2:13Drivers for SQL on Hadoop adoption
- 3:15Actian's SQL on Hadoop announcement
- 4:43Actian's end-to-end platform approach
- 6:06Benchmarking SQL engines on Hadoop
- 8:55Competition and niche roles in SQL on Hadoop
- 9:58Future roadmap and data warehouse trends
- 12:05Early use cases for SQL on Hadoop
- 13:54Actian's community contributions
- 14:58YARN's impact on Hadoop development
- 17:25Benchmark resources and feedback
- 18:08Advice for practitioners
