Databases Reference
In-Depth Information
Disk and Storage Technology
The discussion of hardware architectures in this chapter has thus far centered on system
resources such as CPUs, memory, and I/O subsystems, and noted that parallelism can
take advantage of these resources. An important way to increase hardware performance
is to tune for optimal I/O performance, which includes spreading data across disks and
providing an adequate number of access paths to the data. Since access to disk has the
greatest latency, another focus of I/O tuning is keeping as much data as possible retrieved
from disk in-memory.
On systems designed for performance, disks are often directly attached to nodes or via
a high-speed interconnect such as InfiniBand (in many of Oracle's engineered systems).
Network Attached Storage (NAS) and Storage Area Networks (SAN) provide cost-
effective alternatives but with performance trade-offs. Disks are configured in a variety
of ways for redundancy, eliminating the possibility of single points of disk failure re‐
sulting in loss of access to data.
Disk is commonly deployed in arrays, the industry standard being RAID (Redundant
Array of Inexpensive/Independent Disks). You can use RAID as a part of any of the
configurations we've discussed to provide higher performance and reliability. RAID
disk arrays were introduced in this topic in
Chapter 7
and discussed in the context of
their use in high availability scenarios in
Chapter 11
. Please refer to those chapters for
more information about RAID. In addition, since Oracle Database 10
g
, Automatic
Storage Management (ASM) delivers much of the functionality of a RAID array, such
as striping and mirroring, with a collection of commodity disks. ASM is further de‐
scribed in
Chapter 5
. The storage in an Oracle-engineered system running the Oracle
database is typically mirrored once (e.g., two copies) for normal redundancy and twice
(three copies) for a higher degree of redundancy.
Oracle9
i
first introduced table compression in the Oracle database as a means of de‐
creasing disk storage requirements, primarily in data warehousing. Duplicate values in
a data block are eliminated because values that are duplicated are stored in a symbol
table at the beginning of the block, and all additional occurrences are replaced with a
short reference to the symbol table. Oracle Database 11
g
introduced an Advanced
Compression Option for insert, update, and delete operations important in OLTP op‐
erations. Data compression of three to four times is commonly observed today. Exadata
Storage Servers added support for Hybrid Columnar Compression, described below. In
addition to reducing disk storage, compressed data can also be advantageous for per‐
formance when it allows a set of data to fit entirely into cache (instead of requiring disk
access).
Since disk capacities are constantly growing with newer disks available at lower cost,
many organizations are now storing all relevant data online in disk storage for data
warehousing and business intelligence implementations. Given that disks delivering the
best performance are typically more expensive and of lower capacity, many now deploy
