| 1 |
Pipelined: "assembly line"
-
average query does not have that many steps
-
some operational stages cannot be pipelined (e.g. Sort)
-
one operator is longer than the other (a kind of skew)
-
This is task parallelism in scientific computing
|
| 2 |
Partitioned: "divide and conquer"
-
fairly straight forward to implement
-
"tough" operations are divided, parts run in parallel
-
This is classic data parallelism in scientific computing
|
| 3 |
Combination: "best of both world"
-
several partitions running parallel
-
each partition is, where possible, a short pipeline
|
