Wouldn’t it be great if your networking technology became the default choice among OpenStack users? Two companies are open-sourcing their technologies in hopes of making that happen.
One is Midokura, a known name in software-defined networking (SDN). The startup wants to replicate the success of Ceph, an open source project whose popularity has made it a de facto block storage standard for OpenStack.
Midokura’s competition, though will include some of the people who brought out Ceph.
Akanda is launching today as essentially a spin-out of hosting company DreamHost — the same company that spun out Inktank in 2011 to help popularize Ceph. Akanda sees a similar chance to establish a “material open source” alternative to commercial network stacks for OpenStack, says Simon Anderson, Akanda’s chair and DreamHost’s CEO — and one of the founders of Inktank (which Red Hat acquired in April for about $175 million).
Midokura offers a Layer 2 overlay, and Akanda is a Layer 3 add-on to a Layer 2 overlay, so it’s possible they could co-exist. Both companies are vying for upper-layer work as well, though. What’s interesting is the tactic: Both companies believe an open source option is the only way to capture the OpenStack community’s heart. So they’re bringing their code out in Paris, where the OpenStack Summit kicked off today.
Midokura Opens MidoNet
Midokura’s MidoNet is a Layer 2-4 distributed software overlay for the cloud, targeting IaaS installations. Midokura was an early entrant in network virtualization, back when SDN was just becoming famous, and it’s just started to gain traction with wins at KVH and Zetta.IO, as well as a rumored investment from Fujitsu. Its overlay is going up against big competitors, though — specifically, VMware and its NSX software.
Taking MidoNet open source might sound like a Hail Mary pass. But Midokura — which wasn’t aware of Akanda’s plans as of last week — says it simply saw a wide-open field. No de facto networking option has emerged for driving Neutron, the OpenStack networking project.
More importantly, many of the leading contenders aren’t open, or aren’t open enough for some users’ tastes. Options come down to Open vSwitch and “other,” says Adam Johnson, Midokura’s general manager.
The problem with both is that they’re driven by companies with commercial interest in the results. Open vSwitch is managed by VMware due to its acquisition of Nicira. And other options tend to be put forth by vendors.
“They put just enough input in to satisfy the API spec, but they don’t go much further than that,” Johnson says. “It really is vendor-driven. There really is a problem there. We think it’s hindering OpenStack traction.”
Johnson admits to one defensive aspect of Midokura’s strategy. After some discussion, the startup has decided that OpenStack’s preferred networking option will end up being an open source option. So Midokura is giving in early, joining the open source wave rather than being yet another proprietary or not-quite-open option, he says.
The plan is to have all of MidoNet’s core code available on github today and to launch the midonet.org community site at the same time. (“We’re burning the midnight oil right now to get all the infrastructure in place,” Johnson told us late last week.) MidoNet will be available under an Apache 2.0 license.
Midokura will try to make money the usual way: through a commercial version of MidoNet, called Enterprise MidoNet, augmented with services and support. Enterprise MidoNet would include Midokura’s own user interface (as opposed to using other OpenStack interface options) and code for working with other companies’ environments — a special vSphere-integrated version for mixed vSphere/OpenStack environments, for example.
Pricing for Enterprise MidoNet would be the same as for MidoNet today ($1,899 per host per year) — or $10,000 for a bundled service that includes spinning up hosts for MidoNet and OpenStack.
An open source effort doesn’t mean much without a surrounding community to help advance the code, and that’s where Midokura is going to have to put up some proof. The company does have about 20 community members lined up, most of them vendors. Cumulus Networks, which offers a Linux OS for switches, is on board; Quanta, which makes the kinds of white-box switches that would run MidoNet and/or Cumulus Linux, is in discussions, Johnson says.
But will any of these community members contribute code? They’re not on the hook for that yet.
“In the beginning, it’s fairly lightweight,” Johnson says. “Initially, it’s just showing public support, and as we launch this, we’ll be looking with these community members to see how they can contribute.”
Midokura will govern the community at first, but the community would be able to turn governance over to a different entity later, Johnson says.
Akanda Follows Ceph’s Path
Initially developed at DreamHost, Akanda’s software has been open-source from the start, Anderson says. Like Ceph, Akanda’s software has been getting test-driven by DreamHost (which has been using version 0.9 for about two years) and is now being taken to the open-source community with the version 1.0 launch today.
That was part of the secret with Ceph, Anderson says: The code was battle-hardened at DreamHost. While Akanda has gotten the same treatment, it’s also a step behind, in that it hasn’t had time to build market recognition the way Midokura has.
Akanda provides virtual routing on top of a VMware NSX overlay, along with a network orchestration and management component called RUG. The startup is targeting hosting companies and other service providers, including Tier 2 and 3 telcos. Enterprises are a possibility further down the line, Anderson says.
Akanda’s Version 1.0 adds support for OpenDaylight environments and an API connection to OpenStack’s load balancing as-a-service. Eventually, Akanda wants to also add virtual firewalls, too.
Akanda says all this, including the orchestration, adds up to a network functions virtualization (NFV) play, since it’s targeting carriers. It also seems to pit Akanda against the Layer 4-7 work that Embrane is doing and against Midokura’s Layer 2-4 work — although in the latter case, Midokura has its own Layer 2 overlays whereas Akanda would aim to work with any Layer 2 environment, Anderson says.
What’s missing for Akanda is a wide open source community. “That’s why we’re going out at the OpenStack Summit. We want to start getting people involved,” Anderson says.
For the moment, Akanda is still pretty much a DreamHost project, with DreamHost providing $1.5 million to seed the startup’s launch. “We think that’ll probably take us up to a year of runway to build out a team of maybe seven to 10, mainly [consisting of] engineers and developers,” Anderson says.