MadhavJivrajani/YACS

Yet Another Centralized Scheduler

Yet Another Centralized Scheduler (YACS)

yacs is a tool built to simulate centralized scheduling policies in a distributed setting as a project for the UE18CS322 course: Big Data, at PES University.

Components:

yacs has 3 main components:

1. Master
   • Listener
2. Scheduler
3. Worker

The Scheduler supports 3 scheduling policies:

1. Least Loaded (LL)
2. Round Robin (RR)
3. Random (R)

For more info on these, refer docs for the scheduler.

Getting started

Installation from source

Requirements

• Python 3.6.x

git clone https://github.com/MadhavJivrajani/YACS.git
cd YACS
pip install .

Verify installation

yacs --help

For more info on how to run the master and workers and other related usage info, refer quickstart

If you'd like to build the docs (optional)

cd docs
pip install -r requirements.txt
make html

The docs should now be available in the docs/_build directory.

Using Docker

Run the following commands to start YACS using Docker:

cd docker
docker build -t master . # build image for master
docker build -t w1 -f worker_dockerfile . # build image for a worker

The images can also be pulled from Dockerhub

• Name of master's docker image: aditiahuja/yacs_master
• Name of worker's docker image: aditiahuja/yacs_worker
• Pull the images by running:

docker pull aditiahuja/yacs_master
docker pull aditiahuja/yacs_worker

Running YACS

1. Change directory to yacs/docker and run chmod +x start.sh
2. In the yacs/docker directory, run ./start.sh <scheduling policy> <no. of workers>. For eg. ./start.sh LL 3.
3. Enter the ID and port for each worker and then the number of requests.
4. To inspect the logs, run docker exec -it master bash -c "cat yacs.log".

Design

For the design of the system, please refer to the architecture and the dev docs

Docs

Docs for yacs can be found here

Logging

By default the logs are stored in the directory in which the master / worker is run respectively, to specify a custom path, add an environment variable named YACS_LOGS_PATH to store the logs in this location.

Future work

• Add support for handling faults
  • at Master
  • at Worker
• Partial slots/multiple slot support
• Support for tasks other than map and reduce
• Add support unsynchronized clocks among master and workers (Lamport Clocks?)
• API at Master for retrieving status of jobs

Team members:

1. Madhav Jivrajani
2. Sparsh Temani
3. Aditi Ahuja
4. Ruchika Shashidhara