Future Roadmap of Vectorscan

[markos]

We would like to start a discussion about the future roadmap of Vectorscan.

First a small description of the current status. Vectorscan was born in 2020, after we were contracted by Arm to port Hyperscan to Arm. Intel rejected the Arm patches and the fork Vectorscan was born. Since then, we have been working hard to optimize it, improve the code quality, fix the problematic code and port it to more architectures. Arm has been funding this effort for the past years and even contributed some SVE code to improve performance on such platforms. Unfortunately, this is now reduced to a small maintainance contract which is not enough to further develop of the project. We have many ideas about new features, ports and improvements but all these require time and effort.

At the same time, we have contacted other companies (IBM, RISC-V vendors) and were contacted ourselves by some others vendors who were interested in a port. All were reluctant to fund a porting effort. Having a PR submitted is not enough, even if it works and passes CI, which also requires effort to setup. The reason is that we have been bitten by this in past projects. as we are still left with maintaining that code base, as the original porter might not be working with that company in the future, etc.

So, we have decided not to accept PRs with significant code changes, such as a full port or major refactorings or even new features. We will happily accept PRs for small changes and bugfixes, of course.

For the future, we have 2 options:

1. We split the project in an Open/Community Version with the current feature set, and an Enterprise version, which will include more optimizations, more ports, more platforms that are not included in the Open version, and even extra features that will never be included in the Open version. If required by the refactoring, possibly some more elaborate optimizations will be moved to the Enterprise version, but this will not be done on purpose, we will not cripple the Open version. The enterprise version may be a major refactor and might include features such as ML training, etc.
2. We sell the entire project.

As mentioned, we are a small company but in the past months, we have had to downsize even more, as noone, apart from Arm, has actually paid for any kind of support or development effort for this project. We are not decided yet which way to choose, but the current approach where we do all the work for free is just unsustainable, as much as we like FLOSS as a principle, it does not mean doing the work is for free.

We would appreciate the feedback from Vectorscan users.

The Vectorscan Team
VectorCamp

[koobs]

Spit-balling:

• It'd suck to see VectorScan languish
• It'd rock to see VectorScan keep levelling up (there's heaps of clear value to be created).
• It needs to be funded/sustainable
• There's clearly users (commercially in particular)
• Commercial users need super explicit, easy and fast funding options available to facilitate internal OSS funding conversations.
• Full/complete single-org sponsorship contracts are hard, project/issue/scope based contracts complex too. Targeting flat/simple monthly and/or annual options might be worth persuing as a first foundation.
• Do we know/have relatively clear budgets that can be made explicit, to inform publishing something like github sponsor tiers or similar? If not, figuring out those budgets, even at a basic small/medium/large level would be handy first step and help conversations you have with potential sponsors, or employees of existing organisations / potential sponsors.