Cylon is a fast, scalable distributed memory data parallel library for processing structured data. Cylon implements a set of relational operators to process data. While ”Core Cylon” is implemented using system level C/C++, multiple language interfaces (Python and Java ) are provided to seamlessly integrate with existing applications, enabling both data and AI/ML engineers to invoke data processing operators in a familiar programming language. By default it works with MPI for distributing the applications.

Internally Cylon uses Apache Arrow to represent the data in a column format.

The documentation can be found at https://cylondata.org

Email - cylondata@googlegroups.com

Mailing List - Join

We can use Conda to install PyCylon. At the moment Cylon only works on Linux Systems. The Conda binaries need Ubuntu 16.04 or higher.

conda create -n cylon-0.4.0 -c cylondata pycylon python=3.7
conda activate cylon-0.4.0

Now lets run our first Cylon application inside the Conda environment. The following code creates two DataFrames and joins them.

from pycylon import DataFrame, CylonEnv
from pycylon.net import MPIConfig

df1 = DataFrame([[1, 2, 3], [2, 3, 4]])
df2 = DataFrame([[1, 1, 1], [2, 3, 4]])

# local merge
df3 = df1.merge(right=df2, on=[0, 1])
print("Local Merge")
print(df3)

Now lets run a parallel version of this program. Here if we create n processes (parallelism), n instances of the program will run. They will each load two DataFrames in their memory and do a distributed join among the DataFrames. The results will be created in the parallel processes as well.

from pycylon import DataFrame, CylonEnv
from pycylon.net import MPIConfig
import random

# distributed join
env = CylonEnv(config=MPIConfig())

df1 = DataFrame([random.sample(range(10*env.rank, 15*(env.rank+1)), 5),
                 random.sample(range(10*env.rank, 15*(env.rank+1)), 5)])
df2 = DataFrame([random.sample(range(10*env.rank, 15*(env.rank+1)), 5),
                 random.sample(range(10*env.rank, 15*(env.rank+1)), 5)])
df2.set_index([0], inplace=True)
print("Distributed Join")
df3 = df1.join(other=df2, on=[0], env=env)
print(df3)

You can run the above program in the Conda environment by using the following command. It uses mpirun command with 2 parallel processes.

mpirun -np 2 python <name of your python file>

Refer to the documentation on how to compile Cylon

Compiling on Linux

Cylon uses the Apache Lincense Version 2.0