The Business of IT

Answers from OpenStack leaders

This week I attended the OpenStack Israel 2013 conference. The conference was sold out, with over 400 attendees in a standing-room-only auditorium. As I mentioned earlier, I moderated a panel entitled “If OpenStack is the answer, what was the question?” Here is the video. Below is a summary of the panel’s Q&A.

I want to thank the audience at the event for asking provocative questions, and the panelists:

  • Mark Collier, COO of the OpenStack Foundation
  • Chris Jackson, Manager of Big Cloud Solutions at Rackspace
  • Mac Devine, Director and CTO, Cloud Portfolio, Global Services at IBM
  • Mark McClain, Senior Developer at DreamHost
  • Sivan Barlizy, Product Line Manager, CloudBand Business Unit at Alcatel-Lucent

and, for joining the panel unprepared when I invited him on stage halfway through:

  • Florian Haas, CEO of hastexo.

Earlier I posted some questions I would ask the panelists. Time was short, so not all questions were covered.

Summary of panel discussions

What can be done to improve the operability of OpenStack?

Mark Collier said that The OpenStack Foundation actively listens to users and creates the roadmap based on their input, and operability has been improving: upgrading from Folsom to Grizzly is much easier than previous upgrades. The Foundation encouraged the writing of a book about how to operate OpenStack. We also need to remember that we have hindsight bias: it took many years for Linux to mature. OpenStack is maturing rapidly – but expectations are higher and keep rising. Although there are mostly developers, not operators, contributing to the project, we expect this to change over time as it is more widely deployed.

How does Rackspace keep its public OpenStack cloud consistently running the trunk version two weeks behind the project, without user downtime?

Chris Jackson differentiated between enabling OpenStack adopters to easily deploy and the procedures specific to Rackspace and maintaining compatibility with its legacy SliceHost infrastructure it still operates. But, although the Rackspace tech team shares some of their experiences on their blog, the techniques Rackspace uses to accomplish this particular feat are not shared.

OpenStack’s mission is to encourage an ecosystem of interoperable cloud providers. Why does Rackspace want to encourage that?

Chris Jackson said Rackspace encourages people to be able to stand up OpenStack easily. Their goal is to open up all the ways to build and run a cloud, and Rackspace will differentiate based on service quality. However Fanatical Support will not support just any cloud, only Rackspace’s OpenStack flavor, in order to deliver a consistent quality level.

Why don’t we see an ecosystem of public OpenStack clouds actually forming? Where are the billion dollar budgets backing the creation of these alternatives to AWS?

Rackspace views vendor choice as crucial to OpenStack success, so they sponsor choice by helping telcos build OpenStack clouds. Telcos have billion dollar budgets, and Rackspace has a program that helps them get to public cloud without reinventing the wheel.

Why aren’t we talking about Amazon anymore?

It’s because OpenStack has matured and we have real use cases, real OpenStack deployments, to point to. Rackspace claims that AWS and OpenStack may appear to be similar, but they serve very different functions: AWS gives you cookie-cutter infra resources and tells you how to consume them, while OpenStack gives you the freedom to build for your own custom needs, using any vendor you want, without worrying about any single vendor’s roadmap. Rackspace doesn’t see AWS as competitors. When Rackspace competes with AWS on price it’s just because it’s a necessary evil, to stay relevant in the public cloud market. Public cloud prices are based on the presumption of zero customer loyalty. However, hybrid cloud supports a price premium.

Won’t OpenStack always be playing follow-the-leader with AWS?

Mac Devine pointed out that there is a correlation of the roadmaps of these efforts because AWS blazed the trail. In the early days, OpenStack was playing follower. But from this point, from where we are today, it’s equally important to listen to the ecosystem and implement the features that are requested. Of note is that as AWS matured they began releasing services above the IaaS layer that encourage lock-in. The OpenStack approach is to provide a variety of vendors that can be used to supply services – such as relational database as a service.

Q: What is OpenStack displacing?

VMware should be very concerned. There is no lift-and-shift from VMware to OpenStack. People want an open, no-license model, and providers like Rackspace work with customers to help them transition from VM consumption to cloud consumption. Mac Devine said that even among accounts where VMware is dominant, there is increased interest and commitment to open, lockin-free environments for all new projects. These clients are using OpenStack to extend the features of apps that run in VMware. This will increase the pressure on VMware to be more open. Mac believes that that pressure is one reason why VMware spun off Pivotal Labs, with CloudFoundry inside it, into a separate unit.

How do we get enterprise apps into the cloud, then? What is the roadmap for these apps that were developed without cloud in mind?

There are several things missing in order to be able to do that easily. Bare-metal provisioning, single tenancy, location awareness, standards compliance, integration with management utilities, and network configuration are several examples. Alcatel-Lucent has been leveraging OpenStack as a platform that solves common problems, and adding code around it to provide these features, but these additions are not contributed back to the project. Mac Devine pointed out that legacy apps are still being maintained, but those teams maintaining and developing these legacy apps are already on notice that they’ll need to change the way they design apps. Mac’s division at IBM put almost 600 developers on a project to make DB2 aligned with OpenStack, and he is sure we’ll see traditional OLTP apps being adapted and reshaped for OpenStack gradually. The network is one of the things that is very different in legacy and OS. Each app has its own unique network configuration, a snowflake network. Mac recalled that one client rearchitected their application but didn’t think about the network, and when the new app running in OpenStack needed IP addresses it still took them 10 days to provision those manually.
Florian Haas said it’s silly to think that we will rearchitect legacy apps for the cloud. In terms of getting enterprise apps into OpenStack easily, Florian said we need a check box that says “give me this VM, and make it HA.” “We are about 80% of the way there, which means we only have the other 80% left.”
Chris Jackson pointed out that The Foundation is moving toward the convergence of IaaS and PaaS, so control will be given less to each individual app in the future.

What about OpenStack itself – are the internals HA?

Florian said we’re there already. The message bus, relational DB, the API – these things are done. The hard part that is left is the “I want to make a machine HA.”

The Business of IT

Questions for OpenStack leaders

I’ll be moderating a panel of OpenStack leaders next week at the OpenStack Israel 2013 event in Tel Aviv. Last year’s OpenStack Israel event was excellent, and this year’s gathering looks like it will knock it out of the park. Over 350 people are already registered for the conference, and the two days of technical training following the conference are already full. If you’re in the area you shouldn’t miss this conference. And make sure to introduce yourself to me as well.

In preparation for next week’s event I’ve been formulating tough questions to ask the OpenStack leaders. So far I’ve received great suggestions from Nati Shalom, Randy Bias, and George Reese. Here are some of the questions I’ll put to the panel:

  • How committed are you to ensuring compatibility of OpenStack clouds across different providers? What about between versions?
  • Why is it so difficult to operate OpenStack in practice? How do you plan to address this?
  • Is OpenStack the code or is it the APIs?
  • Public cloud providers are naturally incentivized to differentiate by adding proprietary features to their services, thereby increasing lock-in. Doesn’t this pressure to differentiate conflict with OpenStack’s mission to create multiple, interoperable clouds? And, doesn’t this pressure towards fragmentation undermine the value of OpenStack as an alternative to the Amazon Web Services ecosystem? What do you plan to do about this?
  • What existing systems are being replaced by OpenStack implementations?
  • For the vendors among you (e.g. IBM, HP), how do you plan to make money from OpenStack?
  • Why is OpenStack Foundation membership so expensive? What’s wrong with following the model provided by the Apache Foundation? Is the high barrier to entry in the Foundation really necessary?

Do you have a question you’d like to hear the OpenStack leadership address? Let me know, and I’ll report back on their answers.