Branch	Travis CI	Coverity
master
develop

Fdistdump is a fast, scalable, distributed tool to query Internet Protocol flow record files. The master/slave communication model allows it to run jobs on unlimited number of nodes, including a single local node. Message Passing Interface (MPI) is used as an underlying communication protocol. Its basic features include listing, filtering, sorting, and aggregating records. Fdistdump also implements algorithm to quickly answer Top-N queries (e.g., find the N objects with the highest aggregate values) on a distributed system. fdistdump is a fast, scalable and distributed tool to query Internet Protocol flow files. The master/slave communication model allows it to run jobs on unlimited number of nodes, including a single local node. Among other features, fdistdump allows you to list, sort and aggregate records with the possibility to apply powerful record filter.

It is written in the C language and Message Passing Interface (MPI) is used as an underlying communication protocol. It is currently supported only on Linux, other operating systems are not supported/tested.

• libnf: C interface for processing nfdump files.
• MPI: Message Passing Interface. fdistdump requires implementation supporting standard MPI-2.0 or newer (Open MPI, MPICH2, MPICH3, MVAPICH2, Intel® MPI Library, ...).
• bloom-filter-index (optional): A library for indexing IP addresses in flow records using Bloom filters.

There are two options how to get fdistdump: prebuilt packages and source code.

This is the easiest way of installation, but we current build packages only for Fedora-based distributions. We are using Copr (a Fedora community build service) to create the repositories for several distributions and architectures. Check out the CESNET group page. To install fdistdump, you will need an @CESNET/fdistdump repository and a @CESNET/NEMEA repository, which contains some dependencies (libnf and bloom-filter-index libraries).

The following shell commands show how to install fdistdump from repository on distributions using YUM (e.g., RHEL and CentOS):

yum install yum-plugin-copr
yum copr enable @CESNET/fdistdump
yum copr enable @CESNET/NEMEA
yum install fdistdump-mpich     # version compiled against MPICH
yum install fdistdump-openmpi   # version compiled against Open MPI
module load mpi/mpich-x86_64    # load environment for the MPICH version
module load mpi/openmpi-x86_64  # load environment for the Open MPI version

The following commands show how to install fdistdump from repository on distributions using DNF (e.g., Fedora):

dnf copr enable @CESNET/fdistdump
dnf copr enable @CESNET/NEMEA
dnf install fdistdump-mpich     # version compiled against MPICH
dnf install fdistdump-openmpi   # version compiled against Open MPI
module load mpi/mpich-x86_64    # load environment for the MPICH version
module load mpi/openmpi-x86_64  # load environment for the Open MPI version

If you cannot/do not want to use the prebuilt packages, here is how to compile the source code.

git clone https://github.com/CESNET/fdistdump.git
mkdir fdistdump_build
cd fdistdump_build
cmake ../fdistdump
cmake --build .
sudo make install

• Git: git clone https://github.com/CESNET/fdistdump.git
• Tarballs: https://github.com/CESNET/fdistdump/releases/latest

Fdistdump uses CMake build system. CMake can handle in-place and out-of-place builds, enabling several builds from the same source tree. Thus, if a build directory is removed, the source files remain unaffected.

Out-of-place or out-of-source builds are recommend. An out-of-source build puts the generated files in a completely separate directory, so that source tree is unchanged. You can build multiple variants in separate directories, e.g., fdistdump_release, fdistdump_debug, ...

mkdir fdistdump_debug fdistdump_release
tree -L 2
.
├── fdistdump
│   ├── AUTHORS
│   ├── cmake
│   ├── CMakeLists.txt
│   ├── doc
│   ├── examples
│   ├── extras
│   ├── COPYING
│   ├── pkg
│   ├── README.md
│   ├── src
│   ├── tests
│   └── TODO
├── fdistdump_debug
└── fdistdump_release

Then you can change the current directory and point CMake to the source tree:

cd fdistdump_release
cmake ../fdistdump

In-place or in-source builds are also possible. An in-source build puts the generated files in the source tree:

cd fdistdump
cmake .

The most convenient way to configure CMake is to use tools like ccmake (CMake curses interface) or cmake-gui. However, it is also possible to create a CMake cache entry directly from the command-line using options -D <var>:<type>=<value> or -D <var>=<value>.

Fdistdump-defined variables are:

• ENABLE_BFINDEX=<ON|OFF>: Enables/disables the bloom-filter-index library. Enabled by default.
• EXECUTABLE_SUFFIX:STRING: Appends value to every produced executable. Disabled by default.

Fdistdump-defined variables intended for developers are:

• USE_IWYU=<ON|OFF>: Uses/does not use the include-what-you-use tool. If enabled, the Makefile Generators and the Ninja generator will run this tool along with the compiler and report a warning if the tool reports any problems. Disabled by default.
• USE_CLANG_TIDY=<ON|OFF>: Uses/does not use the clang-tidy tool. If enabled, the Makefile Generators and the Ninja generator will run this tool along with the compiler and report a warning if the tool reports any problems. Disabled by default.

There are also many CMake-defined variables, the most useful are:

• CMAKE_BUILD_TYPE:STRING: Specifies the build type on single-configuration generators. Possible values are Debug, Release, RelWithDebInfo, MinSizeRel. Defaults to Release.
• CMAKE_INSTALL_PREFIX:PATH: Installation directory. Defaults to /usr/local.

If the current directory is the build tree, the you can simply run cmake --build . or make if you generated a standard Unix makefile.

If you generated a standard Unix makefile, you can simply run make install or sudo make install.

One can use the DESTDIR mechanism in order to relocate the whole installation. It is used in order to install software at non-default location:

make DESTDIR=/home/user install

which will install using the installation directory (e.g., /usr/local) prepended with the DESTDIR value which finally gives /home/user/usr/local.

cmake --build . --target package          # archives with binaries
cmake --build . --target package_source   # archives with source
cmake --build . --target package_rpm      # RPMs
cmake --build . --target package_srpm     # source RPMs

fdistdump is an MPI application and is necessary to launch it using one of mpiexec, mpirun, orterun or however your MPI implementation calls the process manager. Before running fdistdump, please make sure that your machine (or cluster) is configured correctly and ready to launch applications using MPI.

This can be tested for example by starting program hostname using MPI launcher. Following should launch two instances of hostname on the local node, therefore local node's DNS name should be printed twice:

mpiexec -n 2 hostname

Following should launch hostname on every specified node. Therefore DNS name of every specified node should be printed to your local console:

mpiexec -host node1,node2,... hostname

If there was a problem running these commands, there is probably a problem with configuration of your cluster (network settings, firewall rules, MPI configuration, ...). For more information about running MPI jobs see documentation for your process manager, e.g. man mpiexec.

MPI allows you to to specify how many processes will be launched on each node. The first process will always be in the role of master, remaining processes will be the slaves. It is mandatory to have exactly one master and at least one slave process, otherwise job won't start. Master process doesn't read any data, so you can launch it on a dedicated node (better option) or it can share a node with slave process.

To run fdistdump on a single node, you should launch two processes. First one, the master, is necessary to manage the slave, print progress and results. The second one, the slave, is there to do all the hard work. It will read and process all the flow files and send the results back to the master. Following example command will print statistic about the source ports in flow_file:

mpiexec -n 2 fdistdump -s srcport flow_file

To run fdistdump on multiple nodes, you should launch one process on each node in your cluster, plus one. One slave process per node and the extra one is for the master. As was stated earlier, you can launch master on a dedicated node, where no flow data is stored, but it can also share node with slave process. Following example command will launch dedicated master on m_node and slaves on sl1_node and sl2_node. Both slaves will read flow_file and send results back to master:

mpiexec -host m_node,sl1_node,sl2_node fdistdump -s srcport flow_file

For complete documentation and more examples see the fdistdump manual page:

man fdistdump