-
Notifications
You must be signed in to change notification settings - Fork 17
Building TileDB
TileDB has been tested on Ubuntu Linux (v.15.10), CentOS Linux (v.7.3.1611) and Mac OS X El Capitan (v.10.11.2), but TileDB should work with any reasonably recent version of Ubuntu, CentOS or Mac OS X. Moreover, it has been tested with g++ v.5.3.0 on Ubuntu, g++ v.4.8.5 on CentOS, and Clang Apple LLVM version 7.3.0 on Mac OS X.
TileDB has minimal dependencies, relying only on the following standard libraries:
Required dependencies:
- zlib (for compression)
- LZ4 (for compression)
- Blosc (for compression)
- Zstandard (for compression)
- OpenSSL (for the MD5 hash)
Optional dependencies:
- MPI (for multi-processing operations)
- OpenMP (for multi-threaded operations)
- Google Test (for code testing)
- Doxygen (for code documentation)
You first need to install three useful programs, cmake, GNU wget and git. Simply run:
sudo apt-get install cmake
sudo apt-get install wget
sudo apt-get install git
Simply run:
sudo apt-get install zlib1g-dev
sudo apt-get install libssl-dev
sudo apt-get install liblz4-dev
For MPI, you can either install mpich with
sudo apt-get install mpich
or OpenMPI with
sudo apt-get install libopenmpi-dev
First clone the Blosc Github repo:
git clone https://github.com/Blosc/c-blosc.git
Then run:
cd c-blosc
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX='/usr'
cmake --build .
ctest
sudo cmake --build . -- target install
First get and unzip the build folder:
wget https://github.com/facebook/zstd/archive/v1.1.0.tar.gz
tar xf v1.1.0.tar.gz
Then install the library:
cd zstd-1.0.0
sudo make install PREFIX='/usr'
First run:
sudo apt-get install libgtest-dev
This command does not create the library - you need to do some manual building. Run the following:
cd /usr/src/gtest
sudo cmake .
sudo make
sudo mv libgtest* /usr/lib/
If you wish to compile the TileDB code documentation, you need Doxygen. Simply run:
sudo apt-get install doxygen
Run:
git clone https://github.com/Intel-HLS/TileDB.git
This will create a directory TileDB in you current working directory, containing the TileDB source files.
Now you are ready to build TileDB. First, enter the TileDB source file directory:
cd <TileDB source file directory path>
Then, create a new directory, say build, and enter this directory:
mkdir build
cd build
Next, run
cmake ..
This automatically finds all TileDB dependencies and creates the Makefile. Subsequently, create the TileDB C library:
make -j X
where X is the practically the level of parallelism for make (e.g., we use 4 on our machine).
To install the (static and shared) libraries, run
sudo make install
This typically installs the TileDB libraries in /usr/local/lib and the header files in /usr/local/include.
To create the examples, run:
make examples -j X
To run the code tests, run:
make check -j X
You can create the Doxygen documentation by running:
make doc
Finally, you can clean up the built files by running:
make clean
Please make sure to read the Compilation Flags and Troubleshooting sections below for some important information you may need depending on your machine and preferences.
You first need to install two useful programs, cmake, GNU wget and git. Simply run:
sudo yum install cmake
sudo yum install wget
sudo yum install git
Simply run:
sudo yum install zlib-devel
sudo yum install openssl-devel
sudo yum install epel-release
sudo yum install lz4-devel
sudo yum install blosc-devel
For MPI, you can either install mpich with
sudo yum install mpich-devel
or OpenMPI with
sudo yum install openmpi-devel
First get and unzip the build folder:
wget https://github.com/facebook/zstd/archive/v1.1.0.tar.gz
tar xf v1.1.0.tar.gz
Then install the library:
cd zstd-1.0.0
sudo make install PREFIX='/usr'
Get the Google Test development release:
wget https://github.com/google/googletest/archive/release-1.7.0.tar.gz
tar xf release-1.7.0.tar.gz
Compile the sources:
cd googletest-release-1.7.0
cmake .
make
Move the header and library files:
sudo mv include/gtest /usr/include/gtest
sudo mv libgtest_main.a libgtest.a /usr/lib/
If you wish to compile the TileDB code documentation, you need Doxygen. Simply run:
sudo yum install doxygen
Run:
git clone https://github.com/Intel-HLS/TileDB.git
This will create a directory TileDB in you current working directory, containing the TileDB source files.
Now you are ready to build TileDB. First, enter the TileDB source file directory:
cd <TileDB source file directory path>
Then, create a new directory, say build, and enter this directory:
mkdir build
cd build
Next, run
cmake ..
This automatically finds all TileDB dependencies and creates the Makefile. Subsequently, create the TileDB C library:
make -j X
where X is the practically the level of parallelism for make (e.g., we use 4 on our machine).
To install the (static and shared) libraries, run
sudo make install
This typically installs the TileDB libraries in /usr/local/lib and the header files in /usr/local/include.
To create the examples, run:
make examples -j X
To run the code tests, run:
make check -j X
You can create the Doxygen documentation by running:
make doc
Finally, you can clean up the built files by running:
make clean
Please make sure to read the Compilation Flags and Troubleshooting sections below for some important information you may need depending on your machine and preferences.
Mac OS X uses Clang as its default C++ compiler, which unfortunately does not support OpenMP. Therefore, the TileDB builder deactivates OpenMP under Mac OS X, which comes at some small performance cost (e.g., internal sorting is not parallel any more when loading unsorted sparse cells into an array).
First, you need to install Homebrew, which is a package manager for Mac OS X. After installing Homebrew, you will be able to install packages to your Mac by running:
brew install <package-name>
You can find the installation guidelines on its website, but an easy way to install Homebrew is by running the following command:
ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"
You also need to install three useful programs, cmake, GNU wget and git. Simply run
brew install cmake
brew install wget
brew install git
Simply run
brew install lzlib
brew install openssl
brew install lz4
For MPI, you can either install mpich with
brew install mpich
or OpenMPI with
brew install openmpi
First clone the Blosc Github repo:
git clone https://github.com/Blosc/c-blosc.git
Then run:
cd c-blosc
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX='/usr'
cmake --build .
ctest
sudo cmake --build . -- target install
First get and unzip the build folder:
wget https://github.com/facebook/zstd/archive/v1.1.0.tar.gz
tar xf v1.1.0.tar.gz
Then install the library:
cd zstd-1.0.0
sudo make install PREFIX='/usr'
Get the Google Test development release:
wget https://github.com/google/googletest/archive/release-1.7.0.tar.gz
tar xf release-1.7.0.tar.gz
Compile the sources:
cd googletest-release-1.7.0
cmake .
make
Move the header and library files:
sudo mv include/gtest /usr/include/gtest
sudo mv libgtest_main.a libgtest.a /usr/lib/
Note: The last commands may not work on your Max OS due to Apple's System Integrity Protection. If that is the case, you need to reboot your system and press cmd+r while it is booting to enter the recovery mode. Then go to Utilities > Terminal from the menu at the top, and run the following commands:
csrutil disable
reboot
This will reboot your system, and then you will be able to move the header and library files as described above.
If you wish to compile the TileDB code documentation, you need Doxygen. Simply run:
brew install doxygen
Run:
git clone https://github.com/Intel-HLS/TileDB.git
This will create a directory TileDB in you current working directory, containing the TileDB source files.
Now you are ready to build TileDB. First, enter the TileDB source file directory:
cd <TileDB source file directory path>
Then, create a new directory, say build, and enter this directory:
mkdir build
cd build
Next, run
cmake ..
This automatically finds all TileDB dependencies and creates the Makefile. Subsequently, create the TileDB C library:
make -j X
where X is the practically the level of parallelism for make (e.g., we use 4 on our machine).
To install the (static and shared) libraries, run
sudo make install
This typically installs the TileDB libraries in /usr/local/lib and the header files in /usr/local/include.
To create the examples, run:
make examples -j X
To run the code tests, run:
make check -j X
You can create the Doxygen documentation by running:
make doc
Finally, you can clean up the built files by running:
make clean
Please make sure to read the Compilation Flags and Troubleshooting sections below for some important information you may need depending on your machine and preferences.
TileDB runs on Lustre. Lustre is a parallel block-level file system (FS) enabling clients from multiple nodes to simultaneously read or write from the mounted directories. It is also compliant with POSIX which means local FS file control functions such as read, write and fcntl work out of the box.
TileDB uses POSIX file locks to ensure consistency. Lustre supports POSIX file locking semantics and exposes local (mount with -o localflock) and cluster (mount with -o flock) level consistent locking. Hence, to run TileDB on Lustre, it must be mounted with either of these options. If concurrent processes are writing to a file on the same node then consider mounting Lustre client with localflock instead of flock. This reduces remote procedure calls to the Lustre metadata server.
You can pass various flags to cmake before building TileDB. Specifically:
-
To change the default installation directory (typically
/usr/local), set-DCMAKE_INSTALL_PREFIX. For instance,cmake -DCMAKE_INSTALL_PREFIX=/usrsets the installation library directory to/usr/liband header directory to/usr/include. Note that you will need to latermakewithsudoin case the installation directory needs proper permissions. -
To control the type of build, set
-DCMAKE_BUILD_TYPE=Debugfor debug mode, andDCMAKE_BUILD_TYPE=Releasefor release mode (the default is release mode if you omit this flag). -
To enable verbose error messages, add flag
-DTILEDB_VERBOSE=1(the default is 0, i.e., no error messages). -
To enable MPI, use flag
-DUSE_MPI=1. -
To enable OpenMP for multi-threading, add flag
-DUSE_OPENMP=1(the default is 0, i.e., OpenMP is disabled). Note that OpenMP is not supported with Clang on Mac OS X and, thus, this flag will be ignored. -
TileDB needs your MAC address for creating certain folder names (this is important if you use TileDB in a parallel file system). The MAC address is typically retrieved through some network interface, e.g., "eth0" for late Ubuntu versions, and "en0" for Mac OS X. If your network interface is not "eth0" for Ubuntu or "en0" for Mac OS X, you need to find which interface carries your MAC address (by running
ifconfigin terminal) and then set it with flag-DMAC_ADDRESS_INTERFACE, e.g.,cmake -DMAC_ADDRESS_INTERFACE=em0. -
You can tune the compression level of GZIP, Zstandard and Blosc by setting flags
-DCOMPRESSION_LEVEL_GZIP,-DCOMPRESSION_LEVEL_ZSTDand-DCOMPRESSION_LEVEL_BLOSC, respectively. If you omit these flags, TileDB will use their respective defaults ones. -
You can enable/disable parallel sort in the case of sparse arrays when you write in unsorted mode through flag
-DUSE_PARALLE_SORT, e.g.,cmake -DUSE_PARALLE_SORT=1enables the parallel sort. Note that parallel sort is disabled by default. -
In case cmake does not find your (already installed) OpenSSL library (this usually happens in Mac OS X), you need to manually specify the OpenSSL root directory with flag
-DOPENSSL_ROOT_DIR. -
In case you have multiple MPI compilers (e.g., both mpich and OpenMPI) and you need to choose a particular one, or in case cmake does not properly find your MPI compiler or library, depending on your machine, you may need to set one or more MPI flags that come with cmake, such as the following (use full paths):
-DMPI_C_COMPILER,-DMPI_CXX_COMPILER,-DMPI_C_INCLUDE_PATH,-DMPI_CXX_INCLUDE_PATH,-DMPI_C_LIBRARIES, and-DMPI_CXX_LIBRARIES. Contact us if you need more help here.
Trouble compiling TileDB with (mpich) MPI enabled in Mac OS X: When building TileDB with make and using mpich, if you see an error message complaining about the gfortran library missing, simply run:
brew reinstall gcc
brew reinstall mpich
This should fix any potential inconsistencies between the gcc and mpich versions.
Trouble running (mpich) MPI programs in Mac OS X: We came across the following problem in Mac OS X when we ran an MPI-based program with mpiexec (with the mpich version of MPI). The program failed with a weird message about the hostname. You need to do the following. First, go to the System Preferences > Sharing settings and make sure you enable Remote Login. Then, set a proper host name by running:
sudo scutil HostName your-host-name
where your-host-name is a name of your choice. Next, add 127.0.0.1 your-host-name to the /etc/hosts/ file. Finally, add your public key to the authorized keys file by running:
cat ~/.ssh/id_rsa.pub >> ~/.ssh/authorized_keys
The above should do the trick.
Trouble running (openmpi) MPI programs in CentOS: When running (OpenMPI) MPI programs with mpirun in CentOS, we came across an error where the program was hanging on MPI_Init and then timed out with an error mentioning hfi_wait_for_device. In this case, try to pass --mca pml ob1 to the mpirun command. This fixed the problem on our machine.