This is THE question that runs through the mind when diving deep on OpenStack. Yes, there is great momentum behind OpenStack. Yes, there is good vendor support for OpenStack. And yes, the community is growing with almost 4500 professionals attending the May 2014 Summit in Atlanta, GA (USA).
Can all this help convince businesses to adopt OpenStack?
Neither the momentum and growing support, nor the growing community and growing number of services with API specifications can convince a business to adopt OpenStack.
Real world deployment stories, lessons from the trenches, and willingness on part of the early adopters to take the stage and share with confidence that OpenStack is ready - is THE way to convince the audience.
And that is happening at the May 2014 Summit in Atlanta!
In this post, we want to review the OpenStack journey and look at some proof points for production readiness. If you need a quick primer on OpenStack, you can read this.
OpenStack was NOT ready in 2012
At F5, our goal is to continue being customer focused and solve complex problems through innovative and robust solutions - but, we also clarify what is not possible. This approach helps us work towards building a trusted advisor relationship with our customers. We want to guide the customer and share information that helps make good decisions. If customers asked F5 in 2012 if they should try OpenStack, our answer would have been a simple NO.
In 2012, OpenStack was rapidly evolving and most of the components, particularly the networking focused areas, needed better design. The original OpenStack concepts were designed to provide alternatives to public-cloud like operations for a private data center. Since most public clouds do not expose underlying networking components, OpenStack Nova (the compute provisioning layer) also embedded networking into its layers. If a customer desperately wanted to make their private data center programmable with existing networking topologies, it would have required serious software development work within the OpenStack environment.
Having said that, some brave souls (E.g. Ebay/PayPal) took the challenge and assembled OpenStack environments, used the ESSEX release and even went public about their production rollouts. These bleeding edge deployments proved you could build a production environment around OpenStack, but it would require money, sweat, blood, and willingness to fail and give up without regrets.
OpenStack was SOMEWHAT ready in 2013
In 2013, the community grew, thanks to the ESSEX early adopters and their testimonials. There was an intermediate release, FOLSOM, which further solidified some of the components. The Networking layer was beginning to get de-coupled from Nova and a multi-node installation - with the Network node, Compute node, and Database node – became possible. Then came the release GRIZZLY. Beginning in 2013 we saw vendors like Canonical and Red Hat starting to complete their deployment tools to ease OpenStack installations. Vendors such as HP, Red Hat, and Mirantis also started to package the open source editions. HP even launched a public cloud based on the GRIZZLY release. This was a significant validation of the OpenStack idea.
However, though OpenStack was rapidly maturing, you still needed enough system knowledge and Python scripting for assembling all the moving parts. In addition, some of the vendor plug-ins didn’t work (hence release stability issue). Vendors promised to fix their plug-in issues. In November of 2013 vendors began delivering on that promise with the HAVANA release of OpenStack. Still, some of the key networking services, like the Load Balancer, were extension ideas with some open source tools and lacked vendor commitments with a good data model. You could not run these network services within existing topologies, and support production loads without bleeding.
Having said that, clearly, momentum was shifting. New success stories at companies like Best Buy were presented at the OpenStack Summit in Portland. Meanwhile, vendors such as MetaCloud began claiming that some of their public cloud customers had reached a tipping point vis a vis scale-to-price thresholds. In other words, these customers were using so much of the computing resources that public clouds no longer had a cost advantage over using OpenStack to build a private cloud.
Gearing up for the first 2014 summit, many were watching with anticipation if it is time for OpenStack to be declared ready for the real world.
OpenStack IS ready in 2014
Though OpenStack still lacks the polish of competitive approaches, it is sufficiently rich in its service offerings and most of the core components are stable for production environments. Moving to OpenStack is now about taking a calculated risk. Yes, you can calculate the cost of investigating it. Yes, you can take the risk -- the chances of success are reasonably high. Why?
First, the list of publicly documented OpenStack rollout success stories continues to grow. F5 is also seeing increased interest within our install base. F5 has been investigating OpenStack for several years. F5’s interest has grown from in-house lab projects to committing to OpenStack integrations. In 2013, F5 joined the OpenStack Foundation and targeted plug-in launches for the May 2014 Summit (we launched!). With a deliberate approach, F5 continued the validation for OpenStack Neutron interoperability with existing topologies and leveraging overlays (E.g. VXLAN) - enabling our customers to avoid reconfiguration of networking services. The support necessary for leveraging existing networking infrastructure is now there in Neutron with the HAVANA release, and getting better in the ICEHOUSE release.
F5 was also validating the Nova compute layer. We integrated with Nova via the BIG-IQ connector to spin up and spin down VMs, running BIG-IP VE instances as well as Application instances. These integrations with BIG-IP and BIG-IQ have been tested in our labs and customer environments, with continued production ready use case validations. F5 has also jointly investigated OpenStack rollouts with some customers - providing guidance and engineering support. In early 2013, we advised some of these customers to plan their move to a public cloud (with BIG-IP VE running in the cloud), and defer looking at OpenStack until later in 2013. Now, some of these customers want to deploy F5 plug-ins into their OpenStack labs and start sharing workloads between the public cloud and their private cloud. Some of these customers are already testing our integrations, and some of them are getting ready for production rollouts.
Second, vendors like Red Hat, Canonical, Mirantis, OpsCode, Persistent, have launched testing tools and programming toolkits to make OpenStack deployment and programming relatively painless. Yes, you need Linux system administration skills and might need some beginner scripting skills, but the tutorials are easy to follow – you can watch a training video and deploy OpenStack in a multi-node environment in a matter of hours.
That's a fair ask. Here's a summary of real world stories for production OpenStack deployments.
Enterprises (E.g. BestBuy, Ebay) delivering Developer self-service and IT efficiency
- Over 15000 Compute instances created and deleted
- 500 active Compute instances, going up to 1500
- Integrated into existing networking topologies
- Developer focused adoption (No tickets!)
- Remove the blame-game (no more "IT caused downtime!")
- Parallel development (Build fast!)
- Reduce cycle time from code complete to code deploy (Deploy fast!)
- IT focuses on automation and monitoring
Providers (E.g. ATT, Comcast, Bluehost) maintain and scale tenant environment without downtime
- Running OpenStack in production in double digit Data Centers globally
- Scale up replaced with Scale out (add more servers to increase capacity)
- Setup an entire infrastructure stack for an event (e.g. NCAA real-time feeds), use it, tear it down once the event is done
- Compute node belongs to the Tenant, but provider can patch the underlying OS with live migration without affecting workload availability
- Use of APIs enables custom live upgrade (this is possible with commercial software too, but now possible in open source software with API customizations) - Upgrade algorithm: HA + 1 - Use a combination of Scale back, rollback - do HA + 1, then live migrate to new node, then validate, then update the taken off node.
My friend and a peer at F5, Nathan Pearce, just posted fresh Day 2 developments from the OpenStack Summit in Atlanta, GA. It has additional proof points for OpenStack production readiness and also what's missing (pay attention to the comments from session presenters).
Customers can invest in OpenStack now. Yes, it will require investment unlike a point-and-click cloud platform deployment experience. But, it will also provide an excellent opportunity to begin transforming IT processes and toolset. F5 is committed to developing OpenStack integrations focused on production ready use cases. F5 can also help customers by leveraging our eco-system (IBM, HP, Red Hat, Mirantis, Persistent etc.) to deploy OpenStack in lab environments. Starting today, customers can also sign-up for using our OpenStack integrations. We invite you to stop by and discuss these topics with F5 - booth E47 at the Summit underway in Atlanta, GA.