OLAP
Learned in SE464.
Online Analytical Processing: the workload BI and data mining systems run.
- Back-end: batch workloads, low concurrency
- Tasks: complex analytics, often ad hoc
- Access patterns: full table scans, big data
Example query: “Over the past three years, in which locations have customers been most responsive to our mailed-to-home coupons?”
Analytics queries involve scanning tables, not using indexes. They must be parallelized across many nodes. The workload is mostly reads, with occasional imports of new data.
To satisfy both OLTP and OLAP needs in one organization, you typically use a data warehouse.
Column-oriented databases are optimized for SQL analytics workloads: storing values of each column contiguously makes scanning and aggregation fast.