Oracle Essentials [Electronic resources] : Oracle Database 10g, 3rd Edition نسخه متنی

11.9 Which Platform Deployment Solution?

In a
world in which there was no limit to the amount of money you could
spend on hardware, you could make a simple decision about the most
appropriate hardware: simply choose the level of throughput and
reliability you need, and go buy it! Unfortunately, we have yet to
discover the location of this kind of world, so your choice of a
hardware solution will often be a compromise.

11.9.1 Platform Comparison

The
most commonly implemented hardware platform for an Oracle server is
the SMP system, which strikes a nice balance between power and price.
SMP systems are popular for the following reasons:

SMP systems offer more and simpler scalability options for the future
than uniprocessor systems.

64-bit processors and operating systems with large memory support
enable SMP systems to handle the needs of very large (even
multi-terabyte) databases.

SMP systems have a single operating system and a single Oracle
instance to manage and maintain, unlike clusters, MPP, or a grid
environment.

Far more applications are certified to run on SMP systems than
clusters, MPP, and grid configurations.

SMP systems are sometimes less expensive than NUMA, clusters, MPP, or
grid configurations in similar CPU configurations because memory and
I/O subsystems are not duplicated to the same degree.

This is not to say that other configurations should not be
considered. Certainly, if scalability demands exceed the capabilities
of SMP machines, clusters or a grid may provide the only viable
solution. Clusters can prove cheaper through use of
"commodity" systems in RAC
configurations. With careful planning and an enterprise-computing
management style, such configurations do provide powerful and highly
available solutions.

NUMA offers an interesting alternative approach. At this point, HP
platforms host some of the largest Oracle based implementations,
scaling to databases in the tens of terabytes of data with more than
1,000 concurrent query processes.

Table 11-2 provides a comparison of the relative
strengths of the different deployment platforms for price,
availability, manageability, and scalability.

You should select a storage technology based on your performance and
recovery requirements and budget. In general, more expensive
solutions offer better performance and more flexible availability
options.

11.9.2 Approaches to Choosing Platforms

When
selecting a solution for deployment, most organizations choose
systems that will meet anticipated performance and scalability needs
for the near future, taking into account management and availability
requirements. However, there are two additional approaches to be
considered.

First is the truism with which we're all
familiarthe longer you wait, the cheaper computer hardware
(and related components) get. According to
Moore's Law, credited by
Intel to Gordon Moore in 1965 (and proven many
times over since then), each chip will double in capacity every 18-24
months, each time providing huge leaps in performance.

This continual reduction in price and increase in performance
characteristics is an ongoing fact of life in the computer hardware
industry. But how can you use this fact in planning deployment
strategies for your organizational system architecture?

Buy what you need, when you need it, and plan for the obsolescence of
hardware by recycling it into the organization when it no longer
meets the needs of an individual application. For example,
today's departmental server may turn into
tomorrow's web listener. As another example, grid
deployment enables you to continue to leverage older hardware as part
of the existing computing solution.

Second, remember to consider the effect of hardware upgrades,
particularly CPU upgrades in non-grid solutions.
SMP systems and nodes require
that all CPUs be identical, so if you upgrade one you will have to
upgrade all of them. At some point the vendor will recommend a new
system anyway because other internal features (e.g., memory and I/O
bus technologies) will have been improved, partly to match the
increased capabilities of the new CPUs.

Grid technology is tempting to consider because new machine types can
be added to the grid as they become available. The self-tuning
aspects of Oracle Database 10g make grid
computing more practical by eliminating difficult manual tuning
efforts that formerly needed to take into account variations in
systems.