Supporting Hardware - Expert Consolidation in Oracle Database 12c - page 81

Unix gurus from the days of Dennis Ritchie, Brian Kernighan, and Ken Thompson, but ultimately their view is focused
on their specific area of expertise. Therefore it is more often than not the database administrator or engineer who can
provide the big picture! Hopefully after you read these sections of the topic, this can be you!
Besides the organizational difficulties just described, there is another reason all the different technologies have
been laid out. When you are thinking about the future hosting platform, it is beneficial to roll out a uniform hardware
platform. Storage tiering is certainly going to stay with us for the foreseeable future. Its main uses will be information
lifecycle management and offering a different cost structure to internal clients. Whether moving data from one tier to
another is going to be an automatic process or manual depends largely on the maturity of the automatic solution and
the comfort level of engineering to release that feature into production.
Flash storage is going to be increasingly used, and there is a potential to design different hardware classes as the
consolidation target. One could think of the platform arrangement described in Table 3-2 .

Table 3-2. Potential Hardware Platforms

Platform designation Characteristics

Gold Rack-mounted servers, using multiple rack units (³ 4U) with lots of DRAM for data

intensive processing. Many fast CPU cores with latest generation processors.
Flash storage could be provided externally via Infiniband or, more traditionally via 8 or
16 GB/s Fiber Channel. Magnetic disk is available in form of Fiber Channel attached
storage array(s).
This platform should primarily be used to more critical production databases with high
I/O demands. If high availability is a concern, then the database might be clustered with
the Real Application Clusters option, hence the requirement to use flash storage outside
the server itself.

Silver This could be the mainstream consolidation platform. The server could use two to four

sockets with reasonably fast CPU cores or dual socket with very fast cores. Total memory
capacity is less than for the gold platform for architectural reasons.
High I/O requirements could be satisfied by PCIe-based SSD as an option. A PCIe SSD can
either store data in a way a directly attached block device does, or it act as a write-through
cache.
In addition to the “bread and butter” production workloads, such a platform could
be used as integration testing platform for the gold servers to save cost. If possible, it
should not be used as a UAT platform for gold servers. Using different architecture and
hardware has never been a great recipe to ensure a smooth testing period—and more
importantly—a smooth production rollout.

Bronze The bronze platform could be aimed at development and integration systems. These are

early in the development cycle, and rapid provisioning of a production clone to debug
a serious issue is more important than the latest and greatest technology, memory, etc.
Another use case for these systems is as repositories for slowly growing data.
There would most likely be no flash storage due to cost considerations.

Table 3-2 's matrix has only taken storage and CPUs into account. There are, of course, many more factors
influencing the final offering from the engineers. The purpose of this section was to encourage you to think out of
the box and include more than the immediately obvious aspect into the mix. The preceding table has deliberately
been kept short. Offering too many options could confuse less technical users of the hosting service, leading to lots of
questions that could be avoided.