FOM Tools provide utilities to study the memory allocation patterns of an application. It has two modes

• Find Obsolete Memory.
• Record memory (de-)allocation patterns

First mode aims to find pages which move to swap and remain there. Such pages contain objects which are not used any longer by an application and as a consequence the memory has become obsolete and could be potentially released.

Obsolete memory tools consist of three parts:

• preprocessing to evaluate the right memory limit for an application. The limit should be such that the application does not start thrashing.
• monitoring utility to track page status and malloc calls
• postprocessing tools to map objects to swapped pages, to plot and merge the data and to make statistics

Requirements:

• Numpy
• Matplotlib (> version 1.2.0)
• compiler with C++11 support
• libunwind
• Cgroup (Memory/Freezer)

Second mode interposes malloc/calloc/realloc and free calls and records time, stacktrace and size. This information is then saved to a record file for post processing. During the postprocessing step four metrics have been calculated.

• lifetime of allocation
• allocation density in unit time
• locality of allocated addresses
• variations in allocation sizes

How to configure and install:

git clone https://github.com/FOM-Tools/FOM-Tools.git
mkdir build
cd build
cmake -DCMAKE_CXX_COMPILER=$(which g++) -DCMAKE_C_COMPILER=$(which gcc) ../FOM-tools -DCMAKE_INSTALL_PREFIX=${PWD}/../install
make
make install

A detailed description about the tool can be found here: https://github.com/FOM-Tools/FOM-Tools/wiki