LinnOS Artifact

This artifact is designed to be run on Chameleon testbed using Jupyter. It will run two end-to-end workflows on a Chameleon instance: baseline and LinnOS (more detail in README). Specifically, it will conduct the following steps:

  1. Process IO traces to obtain Baseline results: a. Acquire an instance with a local SSD array from Chameleon b. Populate those drives with preprepared I/O trace dataset c. Run the SSD replayer and obtain the baseline cdf results
  2. Train the LinnOS ML model and implant the learned weights to LinnOS kernel code : a. Train the LinnOS ML model using the output obtained from running the baseline and save the learned weights
    b. Use header generator to convert saved weight files to LinnOS kernel compatible headers (and put them inside
    LinnOS kernel source code)
  3. Install LinnOS kernel: a. Prepare the config file b. Install the required packages c. Compile and reboot
  4. Using the LinnOS, replay the IO traces to obtain ML results: a. Run the SSD replayer (with LinnOS Kernel enabled) and obtain the linnos-ml cdf results
365 1 June 1, 2021, 2:18 p.m.


  • Levent Toksoz (University of Chicago)
  • Mingzhe Hao (University of Chicago)
  • Nanqinqin Li (University of Chicago)
Launch on Chameleon

Launching this artifact will open it within Chameleon’s shared Jupyter experiment environment, which is accessible to all Chameleon users with an active allocation.