Experimental reproducibility for computer scientists
7th Novembre 2019
Experimental reproducibility is of primordial importance in several fields of research, including computer science. Despite so, it is far from trivial to reliably build software systems that can be tested under challenging environments. Consider a large-scale distributed system (e.g., a key-value store) over a public cloud infrastructure such as Amazon AWS. Its performance (e.g., latency, throughput) vary greatly through time, in unexpected (and sometimes hard to anticipate) manners. System deployers only found out about the unexpected behaviours once the systems are up & running, when perhaps they have to face unexpected service interruptions. This is particularly dauting when measurements done on one day differ greatly from the same ones executed, for instance, after some time by different team of researchers in (apparently) similar conditions. While emulators or software simulations provide partial insights, they clearly lack the benefits of real-life deployments.
This 1-day workshop will address practical aspects of evaluating the reliability and dependability of software systems, covering different facets of experimental reproducibility.
Students will be exposed to tools and best-practices to evaluate real-world systems built on top of de-facto standard echnologies (such as Docker and Google Kubernetes).
It's worth mentining how several major CS conferences have started implementing research reproducibility efforts (check out for examples SOSP'19, PLDI'19) and ongoing efforts are in place since years.
Dr Valerio Schiavoni, IIUN, Université de Neuchâtel.
Prof Adrian Holzer, IMI, Université de Neuchatel.
Prof Christian Cachin, Institute of Computer Science, University of Bern.
The program, as well as the schedule of the talks, can be subject to last-minute changes.
Note: Change of room: A-317 (same building, different floor).
• 08h45-09h00 : welcome and introduction
• 09h00-10h00 : Prof Miguel Matos, U. Lisbon (part 1) - Hang to the helm and sail through a thunderstorm (or how to easily evaluate container-based distributed systems under challenging network conditions)
• 10h00-10h15 : pause
• 10h15-11h15 : Prof Miguel Matos, U. Lisbon (part 2) - Hang to the helm and sail through a thunderstorm (or how to easily evaluate container-based distributed systems under challenging network conditions)
• 11h15-11h30 : coffee break
• 11h30-12h30 : Dr François Vessaz, SBB AG - Reproducibility in mobile applications development and testing
• 12h30-14h00 : meal at the UniNE cantine (provided for registered CUSO members)
• 14h00-15h00 : Dr Arielle Moro, UniNE - Sharing data with the scientific community: opportunities and challenges
• 15h00-15h30 : coffee break
• 15h30-16h30 : Dr Arnaud Legrand, CNRS/INRIA - Toward Reproducible Research in Computer Science
• 16h30 : wrap-up.
Further details will be announced later.
- Pr Miguel Matos, University of Lisbon, Portugal
- Dr François Vessaz, SBB AG, Switzerland
- Dr Arielle Moro, UniNE, Switzerland
- Dr Arnaud Legrand, CNRS/INRIA, France
This 1-day activity will take place the 7th of November 2019 at the University of Neuchatel, UniMail building, Faculty of Science, room A-317.
For details and requests: [email protected]