Skip to content

Latest commit

 

History

History
100 lines (64 loc) · 3.89 KB

INSTALL.md

File metadata and controls

100 lines (64 loc) · 3.89 KB
Raw

Installing this artifact

Prerequisites (for reproducing our experiments)

This artifact makes heave use of Docker and Docker Hub. To reproduce our experiments we assume the user has the following:

  1. A machine with a working installation of Docker.
  2. A good internet connection.

We have tested our docker images on both Linux clients (Ubuntu/Debian and CentOS) as well as a Windows laptop with docker installed.

Running experiments (to reproduce the results reported in our paper)

In addition to our end-to-end toolchain demonstrations (descried below), we have provided routines to reproduce each of our experiments. These reproductions do NOT include running our toolchain on Linux end-to-end (as this takes about 4 hours on our 64 core workstation). Instead, we have opted to provide pre-baked vectors.

Reproducing rq1

See reproduce/rq1/README.md or just run the following:

docker pull jjhenkel/code-vectors-artifact:rq1
docker run -it --rm jjhenkel/code-vectors-artifact:rq1

Reproducing rq2

WARNING: can take up to 4 hours and 6 GB of space. (Runtime on a 8-core desktop machine is about 3 hours.)

See reproduce/rq2/README.md or just run the following:

docker pull jjhenkel/code-vectors-artifact:rq2
docker run -it --rm jjhenkel/code-vectors-artifact:rq2

Reproducing rq3

NOTE: can take up to 1 hour and 2 GB of space.

See reproduce/rq3/README.md or just run the following:

docker pull jjhenkel/code-vectors-artifact:rq3
docker run -it --rm jjhenkel/code-vectors-artifact:rq3

Reproducing rq4

See reproduce/rq4/README.md or just run the following:

docker pull jjhenkel/code-vectors-artifact:rq4
docker run -it --rm jjhenkel/code-vectors-artifact:rq4

Prerequisites (for our end-to-end demo)

This artifact makes heavy use of Docker and Docker Hub. As a prerequisite this artifact assumes the user has the following:

  1. A machine running some Linux distribution. (We've tested on Ubuntu/Debian and CentOS.) We've also done some limited testing of the end-to-end demo on OSX, but support for that platform is still experimental.
  2. A working installation of Make and Git.
  3. A working installation of docker (docker run hello-world produces output without errors).

Installing and running an end-to-end demonstration

Given these prerequisites installing this artifact should be easy. To install follow these steps:

  1. git clone https://github.com/jjhenkel/code-vectors-artifact
  2. cd ./code-vectors-artifact
  3. make end-to-end-redis
  4. make learn-vectors-redis

What make end-to-end-redis is doing

NOTE: can take up to 3 hours and 4 GB of space. (Runtime on a 4-core desktop machine is about an hour.)

If you were able to run the first three commands (listed above) you have done the following:

  1. Used our spec2image tool to create a docker image (locally) containing the redis program's source files and steps to build that program from those source files.
  2. Built that program (in a container) with our custom version of GCC 7.3.0 running our c2ocaml plugin.
  3. Created an OCaml file for each procedure GCC/G++ encountered while building the redis application from source.
  4. Grouped and merged these transformed OCaml files into larger chunks.
  5. Ran (in parallel) our lsee tool on these merged OCaml files to produce abstracted symbolic traces.
  6. Merged the abstracted symbolic traces into a single trace corpus named redis.traces.txt.

That's a lot of cool stuff!

What learn-vectors-redis is doing

IMPORTANT: make sure you run make end-to-end-redis before attempting to learn vectors.

Here's what learn-vectors-redis does:

  1. Uses GloVe to build a vocabulary / cooccurence matrix
  2. Learns word vectors using some default parameters
  3. Runs a demonstration using some Python and the Gensim library