Skip to content

This repository contains a cloudlab profile and install scripts to setup a cloudlab experiment to reproduce paper experiments that run on cloudlab.

Notifications You must be signed in to change notification settings

deeptir18/cornflakes-cloudlab-profile

 
 

Repository files navigation

This is a repository-based Cloudlab profile that creates a Cloudlab cluster ready to run experiments for Cornflakes, a serialization library that offloads data movement into the NIC.

Machines

This profile instantiates 2 or 3 c6525-100g machines or d6515 machines (one as server, the rest as clients); these machines are connected by CX-5 NICs.

Datasets

To reproduce experiments in the paper, we have provided a cloudlab dataset that will be mounted onto the folder /nfs in the server machine; the cloudlab startup script scp's data locally for running experiments.

Setup scripts

The setup scripts (automatically invoked at experiment creation):

  1. Setup cluster of 2-3 machines connected on a 100GB link or 25Gb link. The IP addresses for the experiments will be 192.168.1.1, 192.168.1.2, and 192.168.1.3; the exact interface for the network depends on the machine type (c6525-100g/25g, or d6515).
  2. Mount dataset onto /nfs using dataset parameter provided.
  3. Give machines ssh access to each other (required to run the experiments later).
  4. Install all ubuntu packages (using apt-get) required to build and run Cornflakes.
  5. Install all R packages and python packages needed to run experiments and graph results for reproducibility.
  6. Download protobuf, capnproto and flatbuffers (baselines) into /mydata/packages/ and install locally.
  7. Download specific version of Mellanox drivers into /mydata/packages/ and install locally. Reboot required (manually) after to load new drivers.

About

This repository contains a cloudlab profile and install scripts to setup a cloudlab experiment to reproduce paper experiments that run on cloudlab.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Shell 51.6%
  • Python 46.7%
  • R 1.7%