Databases Reference
In-Depth Information
servers. Suffice it to say, defining a balanced server and storage combination where
Database Server node performance is matched to Exadata Storage Server cells and linked
via a high-speed interconnect (InfiniBand) greatly helps. In addition, Exadata's enabling
of unique in-storage automated Oracle Database optimization techniques, storage in‐
dexes, and the Smart Flash Cache greatly speeds performance and simplifies challenges
IT faces in designing optimally performing solutions. The use of balanced systems, with
high bandwidth interconnects, and the “special sauce” of Exadata Storage Server soft‐
ware have greatly reduced the potential for I/O bottlenecks, which were previously the
leading cause of reduced performance.
Where Oracle's engineered systems are deployed, other storage subsystems are still
sometimes attached as part of an Information Lifecycle Management (ILM) strategy for
archiving data online. Obviously, performance requirements when accessing archived
data are much less demanding.
Oracle and Parallelism
The ability to parallelize operations is one of the most important features of the Very
Large Database (VLDB). Database servers or nodes with multiple CPUs and CPU cores
are the norm today for database servers. Oracle supports parallelism within single
servers and nodes. Oracle supports further parallelism across multiple node configu‐
rations using Oracle Real Application Clusters. Executing a SQL statement in parallel
will consume more of the machine resources—CPU, memory, and disk I/O—but com‐
plete the overall task faster.
Parallelism affects the amount of memory and CPU resources used to execute a given
task in a fairly linear fashion—the more parallel processes used, the more resources
consumed for the composite task. Each parallel execution process has a Program Global
Area (PGA) that consumes memory and performs work. Each parallel execution process
takes its own slice of CPU, but more parallel processes can reduce the total amount of
time spent on disk I/O, which is the place in which bottlenecks can most frequently
appear.
Two types of parallelism are possible within an Oracle Database:
Block-range parallelism
Driven by ranges of database blocks
Partition-based parallelism
Driven by the number of partitions or subpartitions involved in the operation
The following sections describe these types of parallelism.
