Optimizing Aggregate Query Processing in Cloud Data Warehouses - Data Management in Cloud, Grid and P2P Systems - page 10

O.orderstatus of table ORDERS. Hence, we need to scan ORDERS table along
with NationKey-map and CustomerKey-map to process this query.
Figure 5c compares the time taken for processing Query 3. To answer this
query, we need to scan PARTSUPP table, NATION table, NationKey-map, and
SuppKey-map.
Figure 5d compares the time taken for processing Query 4. This query has an
aggregate operation and a filtering constraint on NPK of table ORDERS. Hence,
we scan NationKey-map, CustKey-map, and ORDERS table to process the query.
Figure 5e compares the time taken for processing Query 5. This query is similar
to Query 4, but needs to keep track of ”count” in order to find the ”AVG”.
In the above experiments, MySQL takes 3 inter-machine communications,
while our framework takes only one communication. In MySQL, each node has

(a) Query 1 (b) Query 2

(c) Query 3 (d) Query 4

(e) Query 5

Fig. 5. Performance Comparison of Query 1 to Query 5 of Table 3