SQL/MX 2.x Query Guide (G06.24+, H06.03+)

Parallelism
HP NonStop SQL/MX Query Guide—523728-003
8-24
Influencing Parallel Plans
Right_Child_Seq_Num: ?
Cardinality: 7.2738547E+004
Operator Cost: 5.6920946E-001
Total Cost: 4.6269255E+000
Detail Cost: CPU_TIME: 0.9101162 IO_TIME: 1.7542597
MSG_TIME: 0 IDLETIME: 0.7 PROBES: 1
top_degree_parallelism: 1
bottom_degree_parallelism: 2
top_num_data_streams: 1
bottom_num_data_streams: 2
bottom_partitioning_function: logphys partitioned(grouping, PAPA
with 2 PA(s), log=exactly 1 partition, phys=range partitioned 2
ways on (([260]equiv(TPCD.XMPS.ORDERS.O_ORDERKEY) )))
Seq_Num: 9
Operator: SPLIT_TOP
Left_Child_Seq_Num: 8
Right_Child_Seq_Num: ?
Cardinality: 3.1110000E+003
Operator Cost: 4.4777110E-002
Total Cost: 1.9319278E+000
Detail Cost: CPU_TIME: 0.1957998 IO_TIME: 0.5657918
MSG_TIME: 0 IDLETIME: 0.7 PROBES: 1
top_degree_parallelism: 1
bottom_degree_parallelism: 2
top_num_data_streams: 1
bottom_num_data_streams: 2
bottom_partitioning_function: logphys partitioned(grouping, PAPA
with 2 PA(s), log=exactly 1 partition, phys=range partitioned 2
ways on (([209]equiv(TPCD.XMPS.CUSTOMER.C_CUSTKEY) )))
Influencing Parallel Plans
Several items can affect the plan selected by the optimizer and can determine whether
you get parallelism in plans:
•
Updated statistics for relevant columns. You should update statistics for any
columns appearing in the query. You should also update statistics for subqueries.
For more information about updating statistics, see Section 3, Keeping Statistics
Current.
•
Number of partitions. The number of partitions heavily influences whether the
optimizer chooses a parallel plan. You must have more than one partition to obtain
a parallel plan. The optimizer uses the partitioning scheme to know how to perform
partitioned parallelism.
•
Number of CPUs. For ESP parallelism, the number of ESPs possible depends on
the number of CPUs per cluster.
•
Compiling and executing. You should compile and execute your queries in the
same cluster.
•
Table partitions. Distribute table partitions evenly across physical disks so that your
query can execute in parallel.