Memory hotplug - Problem Statement and Scope.

Problem Statement:

The ability to add and remove memory while the system is up and
running is extremely useful for several reasons. It is an important
feature for data centers and high-end system because it provides
RAS support (reliability, availability, serviceability) by allowing
replacement of failing memory without any down time. Memory hotplug
also allows for dynamic partitioning where memory can be moved from
one partition to the other as the need occurs. Also to accommodate
customers' growth predictions while avoiding the upfront cost the
use of memory hotplug in Instant Capacity on Demand is very valued.
Last but not least the growing area of virtualization were the
hardware resources are distributed among several virtual machines
make heavy use of hotplugging.

There are different levels of difficulty in the removal of memory
depending on the use of the memory one is considering; for example,
user memory is a lot easier to deal with than kernel memory. Also
because of its nature any operation done on a portion of the memory
has potential repercussions on the whole system's stability and
performance.

The Linux memory hotplug community has first focused on the problem
of adding more memory. The code to support memory add is written and
is currently being submitted into mainline. Regression, integration
and performance testing will need to be done.

The more difficult problem of memory removal is still in the
development phase. In order to allow for memory to be removed
one has to first migrate its content to a new memory location
with no or minimal disruption to the overall system. Depending
on the type of memory that needs to be removed this task can be
very challenging.  As the code gets developed regression testing
is crucial; integration and performance testing will be done as
well as the progress in implementation code allows for it.

Up-to-date documentation from a developer and user point-of-view
needs to be written.

Scope:

The scope of this project is to evaluate the stability and
effectiveness of memory hotplug as it applies in various situations
such as 
1) support for Reliability/Availability by providing a mechanism 
to replace defective memory, 
2) support for dynamic partitioning by allowing transfer of a 
physical system resource from partition to another partition in 
order to accommodate work loads needs 
3) support for memory add and remove to meet the need of virtual 
machines (virtualization applications), 
4) support for Instant Capacity on Demand to allow for memory 
physically present in a system to be onlined as needed.

The tasks involved in achieving this include: 
1) contributing to the community efforts to implement memory 
hotplug by submitting code,
2) performing regression testing on a continuous basis to track
the impact on the latest submits to the stability and correctness
of the code, 
3) performing integration testing to verify that the code changes
for memory hotplug do not adversely impact other parts of the 
kernel and vice-versa, 
4) performing performance testing to verify how the memory 
hotplug code affect the overall performance of the kernel 
5) providing documentation for developers to understand
the implementation details and choices and for user to learn how
to use that feature.

It is important to note that all types of memory no matter what
their use (user, DMA, kernel, etc) will be considered as well as
all types of memory architectures (SMP, NUMA).