Database Reference
In-Depth Information
Finally, you should be aware of specific hardware solutions for running Oracle workloads. Exadata has already
been mentioned in the sections above. The other option you have from Oracle is the Oracle Database Appliance
(ODA). Both of these are part of the Engineered Systems initiative by Oracle, that aims at giving customers a balanced
platform for database processing. Both Exadata and the ODA benefit from the fact that their hardware and software
stacks are standardized. Over and above that standardization, Oracle will provide you with patches to the whole
system. In the case of Exadata, the storage tier as well as the database servers will be supplied with regular patches.
Patches to the database running on top of these are also closely related and specifically provided by Oracle. The
benefit of such a combination is a further reduced effort in certifying the whole stack by the engineering department.
In the ideal case, only the patch application needs to be thoroughly tested before it is rolled out. It's also possible to
create your own platform. You can read more about that approach in the following sections.
Consolidation Features in Linux
The Linux platform appears to the author as one of the most dynamic operating systems currently available for
enterprise use. This can be a blessing and a curse at the same time. Engineering departments usually need to spend
time to certify the operating system with the various components that are required to convert industry-standard
hardware into a platform suitable for mass rollout. As you can imagine, this certification requires time, and the time
needed for a certification is exactly proportional to the resources available. You read in the introduction chapter that
resources in all departments, but especially for engineering, are becoming scarce. Quick-release cycles for Linux
kernels do seem counterproductive in this context. This might be a reason why you see enterprise distributions
seemingly using the same kernel (like 2.6.18.x for Red Hat 5.x) even though the change logs indicate hundreds of
backports from the mainline kernel.
One of the open-source ideals, however, is “release small, but release often,” and that will not change. New
features in enterprise-class hardware are often well supported by the Linux distributions. While the mainline kernel
as maintained by Linus Torvalds and other subsystem maintainers moves quickly, the enterprise distributions are
careful not to release too often. Oracle Linux appears to become an exception as the Unbreakable Enterprise Kernel
(“UEK”) is updated at a faster pace. Each new kernel potentially provides new functionality and support for hardware,
which makes it advisable to check for release notes. The following sections explain some interesting features in Linux
that appeared with Oracle Linux/Red Hat Enterprise Linux 6.
Non-Uniform Memory Architecture with Intel X86-64
The Non-Uniform Memory architecture has been mentioned in a number of places in this topic, and this section is
finally providing more information about it. You read earlier that x86-64-based servers have the capability to address
huge amounts of memory compared to systems a few years ago. So what does that mean? In the days of the classic
Symmetrical Multiprocessor (SMP) systems, memory access had to be shared by all processors on a common bus.
The memory controller was located on the Front Side Bus, or FSB. Main memory was attached to a chip called the
Northbridge on the mainboard. Needless to say, this approach did not scale well with increasing numbers of CPUs
and their cores. Additionally, high speed I/O PCI cards could also be attached to the Northbridge, further adding to
the scalability problem.
The memory-access-problem was addressed by moving the memory closer to the processor. New, modern CPUs
have memory controllers on the chip, allowing them to address their local memory at very high speeds. Furthermore,
there is no shared bus between CPUs; all CPU-to-CPU connections in the x86-64-architecture are point-to-point.
In some configurations, however ,an additional hop is required to access remote memory. In such systems, which are
Search WWH ::
Custom Search