One of the greatest business lessons the late Intel CEO Andy Grove gave us is that we have two choices: adapt or die. It’s also a hard message to take. Typically it’s only when something goes really wrong that we agree to change, but in tech that’s usually too late. I believe that’s where we stand with OpenStack*, and it worries me. The opportunities are huge, but we aren’t addressing them fast enough, which puts us on the path to what Andy famously called a strategic inflection point, when change happens far faster than anyone predicts. Intel remains committed to driving OpenStack success, but if we can’t solve OpenStack’s problems soon, all our progress so far will be for nothing.
Here’s an interesting question: If you were to reimagine OpenStack today, what would you do different? I asked a handful of my engineers that and challenged them to look at how they would address OpenStack performance, scaling, security, and complexity issues. This team spent five months analyzing, coding and prototyping to figure out what it would REALLY take to reinvent OpenStack.
First, we decided to treat virtual machine (VM), container and bare metal workloads equally, integrating them in a single scheduler. Then we set our design targets to scale past 5,000 nodes. To support this, decision-making is decentralized with nodes requesting workloads, and state is separated from scheduler, so the latter is not a bottleneck.
We also implemented a Continuous Integration/Continuous Delivery model for upgrades that promotes smaller, incremental updates versus large, semi-annual, disruptive ones. We simplified usability by integrating additional capabilities (think nova-like networking plus) and developed a new interface to match the extended functionality, all leading to much easier installation, configuration, and operation.
Finally, we concluded that security would be mandatory. Network connections should always be encrypted, of course, and SSL is not optional.
Today at the OpenStack Summit in Austin, I demonstrated a project we are calling Cloud Integrated Advanced Orchestrator, or ciao for short. And the results are impressive: with ciao we launched 10,000 Docker containers and 5,000 Fedora VMs in the same 100-node cluster in just one minute. We tried this with Mitaka* using nova, and ciao was multiple times faster! But it’s more than just speed. ciao offers unified networking out of the box between VMs and containers, which is exactly what customers expect.
You can check out ciao for yourself at the Clear Linux Project site. If you are at the OpenStack Summit in Austin, stop by our booth to see ciao in action and talk with engineers who made it happen.
This is just the kind of rapid reinvention OpenStack needs, and it requires breaking so-called rules sometimes. Everyone in the OpenStack community believes in the power of open source; we also must keep an open mind as we look for the best ways to solve challenges, whether in development process, programming language or elsewhere. I invite you to be part of an honest discussion about what needs to be done for OpenStack to truly succeed. We must move faster, we must innovate, and we must be open to all possibilities. The future of OpenStack depends on it.