Comparison
OLAPvsOLTP
OLAP
your query reads three columns across two billion rows and nobody minds if it takes forty seconds.
The workload shape of analytics: few columns, enormous row counts, aggregation rather than point lookup. Columnar storage serves it because it reads only the columns referenced and compresses each one well. Mixing it onto the transactional database is the most common cause of an unexplained latency spike on the write path.
Full entry →OLTP
your queries touch a handful of rows by key, thousands of times a second, and each one has to come back in milliseconds.
The workload shape of an application database: small reads and writes by primary key, high concurrency, low latency per statement. Row-oriented storage and B-tree indexes exist for this shape. Running a analytical scan against the same instance is how one report takes the product down.
Full entry →