Databases Reference
In-Depth Information
Figure 7: Execution time on parallel database environment
Figure 8: Speedup ratio in parallel database environment
recorded the execution traces of the three SQL in each PASS. The decomposition
of execution time is analysed as shown Figure 9 (PASS2) respectively. Comparing
the elapsed time with the cpu time at Figure 9, we find that both are close for View
SQL and Subquery SQL. This means these SQL's are cpu bound, while SETM
SQL is not cpu bound. Most of execution time of SETM query is dominated by
disk write time for creating temporary table such as R_k and RTMP_k. We can
also see that sort time is almost equal for all three SQL's, which represents the cost
of group by aggregation. In PASS2, SETM reuses item combinations in temporary
table R1 on the secondary storage that is generated in PASS1. We replace it with
view or subquery. Then data is transferred directly through memory from PASS1
to PASS2. Figure 9 indicates that PASS2 of those modified SQL queries only read
data from buffer pool. Thus the disk write time of View SQL and Subquery SQL
is almost negligible, although it is dominant for SETM SQL. This analysis clarifies
the problem of SETM and how to cost can be reduced for View and Subquery
SQLs, which is the key to the performance improvement.
7
Summary and Conclusion
The integration of data management and data processing on RDBMS platform
has several potential advantages. The standard such as SQL enables portability
