Release Strategy

Cerowrt has followed multiple release strategies and naming schemes since its inception, and even the primary author finds it all rather confusing. This page is about the current release strategy. As for how to access older builds, please see the Pre-3.3 Releases page.

Goals going forward for the source trees and binaries:

  • Have a clear development/release split
  • Be easily build-able by those with little or no openwrt experience
  • Produce reproducible results
  • Make possible other architecture support
  • Make possible to do at least some (particularly security) in-field updates
  • Make more possible to feed changes upstream
  • Make doing R&D faster and more productive

Naming scheme (binaries)

Future releases will be named after the kernel they are based on. Using ‘codenames’ like OCEAN CITY or Paris will be dropped.

One of the core goals of cerowrt is to track the main kernel trees as closely as possible. This simplifies maintenance and essential R&D by making it easy to develop code on mainstream architectures (x86,x86_64)
and bring the patches ‘right’ over to embedded - and vice versa. It also makes reproducing and fixing bugs much simpler. The kernel is on roughly a 3 month development cycle.

So what we will see are versions that look like

cerowrt-3.3-rc3-X - where ‘X’ is the build number, and is otherwise meaningless. The 3.3-rc3 part mirrors the existing kernel naming scheme, and thus a ‘final’ cerowrt release will probably look like:

cerowrt-3.3.1-X. We will not track the ‘stable’ releases of the kernel too closely unless there is stuff that is explicitly broken that needs to be fixed.

There will be ‘topic’ branches, much like the existing ‘bql’ branch, where some particular problem is being fixed or explored. Rather than confuse people, ‘topic’ branches will go under some other subdirectory.

The biggest problem this scheme induces is that it is in conflict with openwrt’s release schedule.

Field updates

There’s no real plan to do field updates aside from migrating towards something that might work, more often,
at this time.

Naming scheme (repositories)

Because integrating with upstream openwrt has been problematic, requiring frequent rebasing, a given series (3.3 in this example) will end up in its own repositories when its development cycle formally starts. No rebasing will be performed during the formal development cycle. Conflicting commits will be reverted, or merged, instead. (maybe). Commits will be exclusively pulled from the existing git mirrors such as nbd’s, pushed into the current development tree, and patches made available after testing on both the mailing list and in a ‘topic’ branch that people can pull from.

This is modeled on the present-day kernel development process where the middle-tier maintainer asks
the primary maintainer to just pull from a topic branch. It would be my preference from a patch mangling
standpoint for ‘pulls’ from a topic branch rather than commits from the mailing list happen, but we’ll see.

The git repos will try to enforce openwrt’s commit policies as well. (I don’t know what they are and need to find out)

At the start of a new development cycle, cerowrt will be rebased on openwrt head, the patch sets that didn’t make it upstream will be folded in, with the merges and conflicting commits stripped out, and we’ll start again.

We will do continuous integration over that cycle up until something major changes (like a toolchain change),
then freeze, continuing continuous integration in a topic branch until or if it proves out.

I would prefer to keep stuff in git branches rather than explicitly named repos (cerowrt-3.3, ceropackages-3.2), etc, if it weren’t for the rebasing problem and inexperienced git users.

At the moment I plan on explicitly named repos.

Upstream work

The fastest and best way to get patches into openwrt has been to work upstream beyond it.

So periodically core utilities or applications that need an update or surgery will be forked off into the ceropackages repo, worked on, tested, and patches submitted upstream to the application developer. The out of tree package will be maintained out of tree until the new version makes it back to openwrt.

It has generally been too painful to bother submitting updates to core utilities to openwrt’s own upstream,
except for trivial stuff, but we should at least, try harder. A good way to approach this may be to have
explicit ‘patch review’ days on the openwrt patchwork tree. (and to make sure our tested patches are upstream
at that point). It would be a goodness to get more stuff in general out of patchwork and into openwrt that isn’t ‘ours’ as a way of contributing back to the community.

However several core utilities (iproute2, iptables) are very kernel dependent and need to fork off in order
to more closely track their ‘head’ versions. Perhaps a separate ‘openwrt-next’ tree, similar to the ‘net-next’
tree could serve as a staging area for those and for patchwork.

The existing ceropackages ‘application-latest’ convention will be dropped, and the openwrt feeds ‘preference’ option adopted, instead. This minimizes the delta between the next generation of the package and the old package.

Web site

A virtual host will be established somewhere to shorten the urls.

Roadmaps

The existing roadmap and bug lists consist of many man-years worth of work, which is unfunded and un-resourced. Much of that work continues to happen organically, and having the bug lists so public helps to focus our huge herd of cats. Coming up with a better way to track the openwrt buglist would be useful too.

And: coming up with a way to get the herd of cats funded and fed on a regular basis would best.

The existing roadmaps will be cut down to goals that are achievable on a 3 month cycle on the resources we have, so we can at least give people (especially ourselves!) a clue as to what we plan to do.

We will take advantage of opportunities that come up, of course.

In terms of picking a release date, somewhere around the projected arrival of a linux-X.Y.1 seems best.

In terms of picking a ‘start of new development date’, sometime after patches arrive for a new kernel that prove out (in a topic branch), is the trigger point.

Finding a way to reduce the delta between openwrt’s kernel patch set and the device’s specific patch set
should become a higher priority than it has been, in order to further speed up the cycle.

Core feature set

When development came to a standstill back in November, our core feature set was the most advanced set of ideas in the industry, in building a ipv6 capable box that actually worked the way a geek would want things to work.

While that entire feature set now ‘works’ it’s still a little ‘too geeky’, and we have some new requirements in the coming year.

  • Tons of Gui work
  • IPv6 support
  • DHCP-PD support

The core feature set is sufficiently different from openwrt as to cause problems with integration,guis, etc, etc.

Also it would be good to be testing more openwrt-like configurations in the general case.

The core feature set needs to be nailed down for each release and stay nailed down.

Long term support

There still is no plan for long term support! The goal is to do R&D and get new stuff working and pushed upstream!

To edit this page, submit a pull request to the Github repository.
RSS feed

Recent Updates

Oct 20, 2023 Wiki page
What Can I Do About Bufferbloat?
Dec 3, 2022 Wiki page
Codel Wiki
Jun 11, 2022 Wiki page
More about Bufferbloat
Jun 11, 2022 Wiki page
Tests for Bufferbloat
Dec 7, 2021 Wiki page
Getting SQM Running Right

Find us elsewhere

Bufferbloat Mailing Lists
#bufferbloat on Twitter
Google+ group
Archived Bufferbloat pages from the Wayback Machine

Sponsors

Comcast Research Innovation Fund
Nlnet Foundation
Shuttleworth Foundation
GoFundMe

Bufferbloat Related Projects

OpenWrt Project
Congestion Control Blog
Flent Network Test Suite
Sqm-Scripts
The Cake shaper
AQMs in BSD
IETF AQM WG
CeroWrt (where it all started)

Network Performance Related Resources


Jim Gettys' Blog - The chairman of the Fjord
Toke's Blog - Karlstad University's work on bloat
Voip Users Conference - Weekly Videoconference mostly about voip
Candelatech - A wifi testing company that "gets it".