My blog post on dNOS caused some interesting discussions to happen. My thinking was most people, be they open source people or networking people, don’t care about open source network operating systems. I have my theories as to why, and this Twitter thread contains my thoughts:
*Edit: Since deleted twitter thread once found here.
Reactions from others were all over the map. Bill Koss had the first snarky reply, which also rings true:
It is very important if you are on the conference circuit.— WRK (@WR_Koss) March 14, 2018
Mike Dvorkin‘s reply was also right on the money:
Majority of people don't have to worry about networking. And it's an indication of a very good thing.— the dvorkin ⭕ (@the_dvorkin) March 14, 2018
Pete Lumbis tackled it from an interesting angle, indicating the big players already have their own, which is true, and that it’s a hard problem to solve:
hard problem space + limited commercial interest. For big players they already built or bought. Not to mention focus tends to be "let's reinvent things linux already does well"— Pete Lumbis (@PeteCCDE) March 14, 2018
There are other replies on the thread as well, the discussion will hopefully continue on. I also received an email from Bert Vermeulen. Bert had the pleasure of being involved with OpenSwitch and LibreSwitch from the start. He was one of the very few individual contributors to both projects. Bert had some of his own thoughts on why these open source network operating systems never take off.
What killed OpenSwitch was, obviously, it being overly reliant on HPE, which cancelled (the open source side of) the project – and thus taking away all the momentum and 99% of the people, and it duly died. Libreswitch was an effort to Continue without HP, but as Diego told you, there just wasn’t enough interest.
But you can’t blame HPE – Libreswitch also didn’t work, and the newer OpenSwitch codebase is just a do-over which will fail, and ONL has never taken off, and now Stratum will repeat the mistake, and dNOS etc etc.
The reason these all fail is because they’re corporate-driven projects, which inevitably fail to build a community around the project. These companies see value in sharing development resources, i.e. people, from a financial point of view, but are entirely clueless about what makes open source projects work at all.
Bert is right. These projects always come from corporations which have their own agenda. Open Source becomes a tool, to some extent. The real problem is that this can work as long as you have a strong leader at the sponsoring corporation who truly believes in Open Source and can champion it. If that strong leader loses their resolve or leaves, then all bets are off.
However, another reason these projects fail is around the closed hardware ecosystem, stemming from the ASICs in many of these switches and routers:
The reason that all of these open NOS projects are corporate-driven is because the network chips are all so damn closed – that world never caught on to the “open” revolution of the 90s – and the gear itself is out of reach for passionate hobbyists to play with.
Bert offered some hope to me, however:
Where the gear is in reach of hobbyists – SOHO routers – an open source community has indeed formed: OpenWRT, and it’s a roaring success. The project, for all its warts, now supports broadly every damn thing in scope, and a lot (if not most) products in this market ship with a vendor-supplied version of OpenWRT. That’s the kind of success these corporate-driver NOS projects are chasing, but will never reach.
OpenWRT is an amazing project, and it has done some great things over the years. I remember first using OpenWRT over 15 years ago, and it is still going strong! OpenWRT is an example of an open source network operating system which grew from community roots, and has built a strong contributor base, and is now used by manufacturers of wired and wireless products. It gives me hope that perhaps in the future there can be a datacenter network operating system which can trace it’s roots to a strong open source community.