load shedding for aggregation queries, ICDE2004

B. Babcock, M. Datar, R. Motwani, "Load Shedding for aggregation Queries over Data Streams", ICDE 2004

Problem:
Which tuples to drop, and where in the query plan to drop them, so that the degree of inaccuracy in the query answers introduced as a result of load shedding is minimized.
Model: a query made by multiple operators that ends with an aggregate operator.
Solution:
Seen as an optimization problem: minimizing inaccuracy in query answers, subject to the constraint that system throughput must match or exceed the data input rate.

Uses Statistics Manager to keep track of mean of variance of the values in the input tuples.
QoS: add weights on the queries and use it to compute the relative error introduced by the load shedder.
Questions:
1. How is the load shedding done?