Openstack Summit, Vancouver – Day 2 (May 2015)
Mark Collier, COO at Openstack starts the day going a little deeper into Openstack’s architecture than on Day 1 explaining to the audience that the platform not only supports the major hypervisors, but also bare-metal implementations.
This fact leading nicely onto the first of the customer evidence sessions with Collier introducing James Penick, Cloud Architect at Yahoo, to the stage. Yahoo, an Openstack super-user, has been investing in Openstack for some time.
Referring to themselves as a mega-scale architecture, they have servers in the 100’s of thousands. Managing theme required the development of bespoke tools that was beyond Openstacks initial capabilities: they modified Nova drivers to run it on top of their existing, bare-metal infrastructure. Currently, Yahoo has 10’s of thousands of servers being managed by Openstack with a goal to have all of the aforementioned by the end of the year.
Zack Rosen, CEO of Pantheon, up next representing an organization supporting 400,000 PHP environments. Rosen, explains that solving the random nature of Internet traffic–unpredictable spikes of 10x, often 100x in demand–is difficult without platform agility. Adding to the complexity, at Pantheon they must also ensure isolation between environments within a shared infrastructure.
“Having each customer on their own cluster is wasteful, as most don’t need that much hardware, and some customers have 2000 sites, which won’t even fit in a single cluster.” he said.
Incredibly, Pantheon has only four DevOps staff to run all of this. Possible because they run on a Rackspace Openstack implementation with containers that can rapidly scale up and down as need.
Following Rosen is Adrian Otto, Distinguished Architect at Rackspace, Leader of the Openstack project, Magnum – an API service developed by the OpenStack Containers Team for OpenStack to make container management tools such as Docker and Kubernetes available as first class resources in OpenStack.
While containers have been supported in Openstack for some time (Openstack Nova), initially, containers were treated much like VM’s.
Collier then introduces the Community App Catalogue (http://apps.openstack.org), which opens up today for community contributions. In the Kilo release OpenStack added the ability to import community-generated packages that have been developed by other Openstack community members. Craig Peters, Product Manager at Mirantis, demonstrated the deployment of an application fmorthe App Catalogue showing how OpenStack operators can import the app package into ones own catalogue and how that deployment can also obtain the application dependencies – for example, if the application package requires Docker, it will also import the Docker container packed into the operators catalogue.
The packages themselves, referred to as Murano packages, are available once the Openstack administrator has implemented. On explaining the importance of this project, “the applications are really the key” said Peters. A sentiment familiar to anyone who’s worked with F5 over the last decade.
Up next, Google Cloud Solutions Architect, Sandeep Parikh. Parikh began with a few facts about Google Compute. He explained that Google generates about 2 billion containers every week. The equates to a lot of knowledge about container management, which they then turned into Kubernetes, and released to the world. Parikh’s believes that Openstack is the platform to bring Kubernetes to the world.
To demonstrate, Sandeep and Adrian Otto (Racksapce) setup and environment running Kubernetes on Google compute, and Rackspace OpenStack. As both environments were running Kubernetes, they can deploy the exact same application, however complex, on both the Rackspace and Google Compute environments.
VP and General Manager of Intel’s Open Source Technology Center, Imad Sousou, took the stage next explaining that, in regard to OpenStack, there are still a lot of skeptics out there and there’s still plenty of work to do to make it an easily deployable, robust platform.
“In order to get there we are going to need a lot of focus.” Said Sousou, “Take Linux, for example, it took ten years to get Linux where it is today. We still have a way to go to get Openstack to that level”.
Intel are working on a lighter-footprint Linux operating system named Clear with a goal of spinning up systems and containers in 200 milliseconds.
Following Sousou, we have Subbu Allamaraju Chief Engineer, Cloud at eBay. eBay run one of the largest Hadoop clusters in the world producing 2million metrics a second to their analytics system. Everything at eBay is deployed through automation allowing them 150 service deployments a day.
eBay’s jounrey to Openstack started in the Summer of 2012 when they deployed OpenStack Essex to manage approximately 300 servers. comparing that to now, Spring 2015, they are running OpenStack Havana to manage the following infrastructure:
- >12,000 hypervisors
- >300,000 cores
- 10+ availability zones
- 15+ virtual private clouds
- 1.6 PB of block storage provisioned through Openstack Cinder
- 100% KVM
- 100% OVS
Other than OpenStack Keystone, for managing identity/access, eBay has no service spanning an availability zone–a decision made for resilience.
“Do you believe in Openstack”, asks Allamaraju having shared the metrics above. He then went on to share more data about the services the are running on OpenStack:
- 100% of PayPal web/mid tier
- 20% of eBay web/mid
- 100% Dev/Test workloads
“The developers love Openstack” says Allamaraju. “They can get resources via an API call whenever they need it.” He then went on to say that Openstack is not cloud, it’s a platform controller of which eBay still had to implement solutions for:
- Scale Out
Allamaraju closes out by saying that the Openstack community needs to raise the operation bar of the core of Openstack. He cited comments from Openstack Operator sessions about some of the scaling issues of Openstack services–Heat, for example.
After a short video reminding the audience that Openstack started out as an idea between Rackspace and NASA, that has since evolved to solve the same scale and management demands for many others, we welcome Mark McLoughlin, OpenStack Technical Director, RedHat.
Mark explained the three phases of RedHat’s approach to OpenStack:
- Participation: they look for community. Taking their RedHat badges off and participating in the OpenStack community as equals to their customers and partners
- Deliver: ensuring they build a robust platform that stays close to the Openstack code base without sacrificing stability
- Engage: working closely with their customers to ensure success.
McLaughlin citing critical situations like recent security vulnerability, Heartbleed, as an example of why support models are important, even with open-source solutions like OpenStack.
Last comments from Collier, thanking the sponsors again, and the introduction of the final morning speaker, Jonathon Chang , IT Chief Engineer, NASA JPL (Jet Propulsion Laboratory).
Chang shared the history of NASA’s space exploration, details of the experiments being conducted today, and work being done to get man to Mars. His explanation of projects like the Low Density Supersonic Decelerator, NASA LDSD, and theAsteroid Redirect Mission, kept the audience on the edges of their seats.
Chang then explained the immense volumes of data, and processing of that data, that occurred from each individual experiment.
This was the right audience for NASA, and I’m sure I heard a few gents talking in Klingon afterwards, but, sadly, it was somewhat anticlimactic to return to their OpenStack architecture that keeps all their systems running and accessible.
There were two main themes today:
- Containerization: showing the audience that, contrary to some media coverage, containers are not a threat to OpenStack. They are a well embraced part of the strategy
- Improvements for Operators: as per the comments yesterday from Mark Interrante, SVP Engineering, HP Cloud, its clear that the Operator needs focus and that OpenStack needs to be easier to use. A strategy that must start at the core.