System Story: OpenDSA eTextbook System

System Information

Initial Project Conception

OpenDSA began in early summer of 2011 as a collaboration between Cliff Shaffer and Ville Karavirta while Shaffer was on an extended visit to the research group of Lauri Malmi at Aalto University in Finland. At that point, Shaffer had been working for several years on the AlgoViz project, which promoted greater use of Algorithm Visualization within the CS education community. A key finding from AlgoViz was that instructors find it easier to adopt complete units of instruction rather than individual visualizations. For example, there are many variations on Quicksort. Given access to a high-quality Quicksort visualization, there is a strong chance that it won't match the code or approach presented in the rest of the course instructional materials, making it harder for the instructor to integrate and for the students to understand. Better is to give instructors the ability to adopt a full Quicksort presentation, or a chapter on sorting, or a full data structures textbook.

Karavirta created the JavaScript Algorithm Visualization (JSAV) library during Summer 2011 while Shaffer created demonstration visualizations and associated explanatory text. They worked extensively with Tom Naps of University of Wisconsin-Oshkosh, who had a vision for a community-centered, open-source eTextbook project. This led to using the name "OpenDSA" for the project. Shaffer had recovered the publication rights to a data structures textbook that he published in the 1990's, and so had a large base of prose to work from. One of the first decisions that had to be made was how to author content. After conversations with Brad Miller of Runestone Academy, we adopted the Sphinx compiler and ReStructuredText (RST), the documentation language of the Python community. What made RST attractive to OpenDSA is the ability to extend the language through implementing "directives" in Python, and the openness of the Sphinx compilation process. This gave us a rich authoring tool that could also integrate non-RST elements (like visualizations and exercises) into the compiled HTML files.

This gave us all the building blocks for the eventual project: RST-based prose integrated with JSAV-based visualizations and exercises compiled into HTML pages that could be served to users. OpenDSA's fundamental conception is that it provides infrastructure and a collection of materials, which allows instructors to pick-and-choose from the content to create their own eTextbook. The content selections are saved in configuration files for future use.

The initial server for managing student logins, serving content, and collecting scores was written by Eric Fouh in Django, with the initial client-side framework written by Daniel Breakiron. The first use of OpenDSA to support a course was for a chapter on sorting for a junior-level data structures course at Virginia tech during Fall 2012. One of the first lessons learned was that while good prose and visualizations are important (and students strongly prefer visual presentations of algorithms and data structures), the interactive exercises are the single most important thing in any eTextbook. Most of the early exercises used by OpenDSA were rewritten versions of TRAKLA exercises, contributed to OpenDSA by developers at Aalto University. OpenDSA was used as the full textbook for Virginia Tech's junior-level data structures course for the first time in Fall 2013, with full textbook prose, extensive visualizations, and automatically graded homework exercises for the full semester.

Since this time, use of OpenDSA has grown to be used by dozens of institutions around the world, with full textbook support for course in Data Structures (primarily written by Shaffer), senior-level algorithms (primarily written by Shaffer), programming languages (primarily written by David Furcy and Tom Naps at University of Wisconsin-Oshkosh), and formal languages and automata (primarily written by Mostafa Mohammed and Cliff Shaffer at Virginia Tech, 2019-2023). Visualizations and exercises have been developed over the years by nearly 100 undergraduate and graduate students at a number of universities. By 2020, in a typical semester OpenDSA is used in classes at around 15-20 institutions, mostly in the US but also around the world.

Major Development Milestones

While the original OpenDSA server was written in Django in 2012, we recognized the need to move to a system that could better interoperate with other systems. A major goal in making a change was to break the notion that the OpenDSA server would supply LMS features like logins and maintaining scores, that these should be offloaded to a real LMS. During Summer 2015, we attempted to develop a new server based on the OpenEdX project. By the end of the summer we decided that this approach would not work for us. At that time we decided to adopt Ruby/Rails as our server implementation. The OpenEdX effort and then the initial Ruby/Rails server implementation were led by Hossam Shahin. Equally important, we adopted the LTI protocol for communicating score information between client, server, and LMS. LTI not only allows OpenDSA to communicate with an LMS, it also allows use of 3rd-party exercises that support LTI. At about that time, Virginia Tech adopted Canvas as the institution's LMS. Thus our initial Ruby/Rails server implementation compiled books and made them available within Canvas as Canvas module, with scores communicated between the client-side runtime framework and Canvas (though the OpenDSA server) via the LTI protocol. Additionally, OpenDSA can support some integration with any LMS that supports the LTI protocol, as we also expose all exercises and modules as LTI tools. One weakness of Canvas is that only a single scorable entity can appear on a Canvas module. Initially this meant artificially breaking up what might naturally appear on a page as a "topic" or "section". In 2018 Jackson Wonderley reimplemented the LTI support so that OpenDSA module pages report exercise scores to the server, which in turn aggregates the values of various exercises in the module appropriately and reports them to the LMS. In this sense, the OpenDSA server changed from being just an LTI Tool Provider to also an LTI Tool Consumer. This basic server configuration has been in place since that time.

While we have always aspired to have the OpenDSA infrastructure be easy to install locally by developers and institutions that want to use it, it has historically been fairly difficult to install the OpenDSA server. In 2021, Alex Hicks and Patrick Sullivan implemented a Docker containerized version of the system. Those who only want to develop content can easily install this, while those who want to run a full server stack capable of serving students in courses can also do a relatively easy installation.

Funding and Support

Resources to operate and expand OpenDSA has come over the years in three forms.

  1. OpenDSA has been funded by a series of NSF grants. The first NSF-TUES program grant in June 2012 had PIs Cliff Shaffer of Virginia Tech, Tom Naps of University of Wisconsin-Oshkosh, and Richard Baraniuk of Rice University, with the VT share directly supporting OpenDSA infrastructure development with $125,000. An EAGER program grant from January 2013 to December 2015 supported collaboration between Virginia Tech (Cliff Shaffer, $67,000), UW-Madison (S. Puntambekar), and Auburn University (Hari Narayanan) to work with Aalto University. A major NSF IUSE grant for $1,000,000 supported major OpenDSA development from 2016-2019. While SPLICE grant funding from 2017-2025 does not directly support OpenDSA, it does affect bringing OpenDSA infrastructure in compliance with community standards such as PEML, ProgSnap, and Caliper.
  2. As an open-source project, OpenDSA has had tremendous success at getting volunteers to help over the years. About 100 people, mostly students doing Independent Study or volunteer work, have contributed content in the form of individual visualizations or exercises. The general model has been that paid graduate students build and maintain core functionality, faculty design modules, chapters, or books, and students are recruited to volunteer or get credit to implement visualizations and exercises.
  3. Starting in 2021, the Department of Computer Science at Virginia Tech has contributed a small amount of professional staff time to do basic system maintenance, in recognition of OpenDSA's role as supplying basic instructional infrastructure. In 2023, CS@VT agreed to support OpenDSA directly with 1/4 of an instructor position to do core software development.