Query Processing in Sensor Networks
Yong Yao and J. E. Gehrke,
CIDR'03.  

First Biennial Conference on Innovative Data Systems Research
January 5-8, Asilomar, CA
CFP: "Such papers are typically speculative or are system
evaluations. The visionary papers usually lack rigorous frameworks,
simulations of performance, or prototype implementations. The
prototype descriptions generally are a detailed report on successes
and mistakes....its goal is to air radically new ideas."

TAG came out at OSDI December 2002.

Proposal: distinct query proxy layer in between network and application
layers.  Examined in the following contexts:
(1) in-network aggregation
(2) interaction of in-network aggregation with the wireless routing
protocol
(3) distributed query processing

Comparison to TAG/TinyDB
- DURATION p and EVERY q vs. 
epochs are specified with SAMPLE INTERVAL x s FOR y s [s->sec].
- Here, the process of computing a query answer is called a round.
- They leave event-based queries for future work, whereas TinyDB is
starting to look at them.  It is unclear if TinyDB has them actually
implemented.

In-network aggregation
- direct delivery
- packet merging
- partial aggregation

Synchronization
- straw man: incremental time slot algorithm
- "our approach": notification packets

Modifications to AODV
Maintenance:
- local repair
- bunch repair

Query Plans
- flow blocks
- query optimization
  - group by
  - joins

declarative queries: SELECT FROM WHERE

aggregates can be monotonic (COUNT), exemplary (MIN), or summary (AVG).