There is a need for an open-access, free-for-authors, free-for-readers, high-quality journal for systems research. This repository contains a proposal for such a journal.
JSYS seeks to promote high-quality, open-access research on computer systems. JSYS will never charge readers or authors. All research published in JSYS will be available as open-access from the day it is accepted for publication. JSYS will not impose page limits on accepted publications.
- Top venues like SOSP/OSDI and Eurosys don't do revisions: good work that is not perfect gets rejected and bounced around (wastes both author and reviewer cycles).
- Artificial constraints like being single track limit the number of papers accepted and presented.
- Coupling publication and conference attendance is unfortunate, shutting out authors from countries like Iran and authors who simply don't have the money to travel
- Conference papers also have an unfortunate limit on number of pages both in the submission version and in the final camera-ready version. Some conferences require authors to pay significant amount of money (e.g., $150) per extra page in the camera-ready version.
These problems would all be solved by moving to a journal format. The quality would remain high, similar to SOSP/OSDI and Eurosys.
Conferences would still remain, but they would be explicitly for networking. Authors can publish without attending the conference.
To ensure the open-access mission of the journal is carried out properly, it must be governed by an independent board. Existing journals such as Transactions on Computing Systems are all governed by organizations such as ACM, which have a different mission and set of values. A number of conferences, such as Neurips and RSS, are governed by their own independent foundations.
JSYS would work in a manner similar to PVLDB. Authors can submit to the journal on the first of every month. By the first of the following month, the authors will receive a decision: accept, revise, or reject. A revise decision means the authors can submit a revised manuscript 1--3 months after getting the decision.
Continuous Interaction. During the one month of review, there will be continuous interaction between the authors and reviewers via HotCRP (in a double blind manner). In this way, the need for rebuttal is eliminated. The continuous, anonymized communication helps clears mis-understandings.
Non-confrontational review. The goal of all reviewers to identify what needs to be done to get the paper to a publishable state. There are no constraints such as acceptance rate. The authors can then choose whether to perform the work identified by the reviewers to publish in this journal.
We hope to include senior graduate students as reviewers in the journal. The reviewer set will also include established researchers, to ensure the grad-student reviewers are calibrated.
We think there are many benefits to including senior graduate students as reviewers: they have more time than faculty to do deeper reviews, they are more enthusiastic about the review process, and they are more open to learning how to write good reviews. We also suspect they are less likely to turn in late or extremely short reviews. There is not a significant difference in review quality between a senior grad student and someone who has just graduated.
It will take time before the journal becomes "prestigious". Several new conferences such as NSDI and Eurosys significantly increased in quality since their founding; we expect it will be the same with the journal. We expect it will be five to seven years before the journal is considered "top".
The journal will rely on volunteer work, similar to most computer science conferences. It is not expensive to host the papers published in the journal. The journal will not offer type-setting or copy-editing: the authors will do this themselves (this is common in computer science conferences). In this manner, the journal takes advantage of characteristics unique to computer science research.
Similar journals such as the Journal of Machine Learning Research (JMLR) have been open access for over 20 years without charging readers or authors. The publisher shares the biggest expense is filing taxes.
A number of problems may arise in this model. Dave Levin provided a list of thoughtful questions (which we originally came up in the context of reviewing in security):
- Will there a lack of expert reviewers in this model? Conferences attract expert reviewers; at least initially, it may be hard to find such reviewers for this journal.
- Will reviewers ask for a lot of additional experiments for publication? This has the potential to move the paper away from the vision of the authors, and add (perhaps un-necessary) extra work for the authors.
- Will the lack of in-person PC meeting lead to large variability in quality of accepted papers?
- Will this model encourage lots of low-quality submissions?
We believe these questions can be answered by collecting data during the review process. For example, we will ask authors to rate reviews. We will also collect the self-reported expertise metric from each reviewer.
- Forming a steering committee
- Finalizing details of journal
- Getting the journal registered? Get an ISSN?
We've setup the #jsys channel in the Systems slack to discuss this idea. Please join the systems slack here, and look for the #jsys channel.
- Journal of Machine Learning Research (open-access for over 20 years!)
- Journal of Object Technology
- Compositionality