With more people volunteering to give talks than we had time for, the user group had voted their selections for the packed agenda. The main 3 talks covered 30 minutes each and 10 minute (or so) slots for lightning talks.
Over 60 people attended and the room was full with standing room only. Photos of the event are available here. Recording of the sessions is available at https://cast.switch.ch/vod/channels/2i5k459xe3.
OpenStack HA – Florian Haas, Hastexo
Florian started us off with a rapid overview of approaches for high availability in OpenStack. While the talk was ‘insanely short’, he explained the different types of components in OpenStack deployment and how to make these redundant. The talk is the base for the open that is proposed for the summit in Portland during April so those attending can vote to hear it there.
High availability implementations of MySQL, AMQP and API services were presented based on the Pacemaker suite which while it is not the most user friendly of packages, has become the defacto standard for these implementations. The good news is that the OpenStack implementations can be performed by copy/paste of reference configurations or using Puppet/Chef manifests.
High availability of guests has been a topic for intense discussion at past summits. The cloud providers based on scale out application architectures do not need these features but the providers running enterprise workloads would like functionality such as ‘restart this VM on a different hypervisor if the hypervisor goes down’. While there were some tricks in Folsom which could be used, in Grizzly there is now a ‘node evacuate’ function which performs this operation when the hypervisor is down. Some scripting is still required though as this is not automatic and the choice needs to be made between using nova migrate (if the hypervisor is running) and nova evacuate (if it is down). More work expected in the Havana release…
A couple of features may make the cut before the feature freeze for Grizzly next week. VM ensembles ensures guests providing redundancy are put on different hypervisors to avoid single points of failure. Libvirt watchdog support will restart VMs if they are blocked.
Ceilometer – Lucas Graf, Toni Zehnder – ICCLab
When OpenStack was started, there was no billing functionality included. Ceilometer has been developed over the past year to perform metering (who is using what) and is being gradually extended to cover monitoring (what is working and how hard is it working).
With the Grizzly release, ceilometer will be incubated and integrated into the Horizon dashboard. A number of projects such as healthnmon and synaps are aligning their architectures to benefit from ceilometer and avoid gathering this data more than once. In the Havana release, monitoring will be included including a hardware collector developed by ICCLab to cover IPMI and S.M.A.R.T. monitoring.
It was particularly impressive to see the two presenters who have only been working on OpenStack for 4 months contributing to the community and mastering the topics, despite that the beer and pizza arrived during their talk (thanks to ICCLab and Rackspace)
Heat – Muharem Hrnjadovic – Rackspace
Muharem presented Heat, the orchestration component in OpenStack which is currently in inclubation in the Grizzly release. This provides a template system for creating applications similar to Amazon CloudFormations where you describe the different components of the application, define which parts can be scaled up or down and deploy to an OpenStack cloud.
Although Heat has only been developed during the past year, Muharem demonstrated its stability by running a live demo, creating a wordpress site smoothly.
OpenStack on OpenStack – Paul Voccio – Rackspace
Rackspace needed an easy way to spin up new OpenStack instances for development, test and production. To do this, they run OpenStack instances on top of OpenStack! Cutting an initial golden image, using load balancers to retire old instances and bring online new ones, they’re using approaches pioneered in web application development and applying them to infrastructure.
Snabb Switch – Luke Gorrie
Luke gave a talk without slides covering the motivations behind developing Snabb Switch (http://www.snabb.co/snabbswitch/). Snabb Switch provides an ethernet switch in user space, built on Open Source and implemented outside of the kernel which leads to improved performance and ease of implementation (much of it is written in Lua). The switch itself can therefore be easily extended in a similar way to writing macros for editors.
Cloud Foundry – Christof Marti
The current status of the Cloud Foundry PaaS was presented covering frameworks such as Rails, Spring and Java along with services such as MySQL and MQ. The applications can be deployed on multiple cloud providers. Recently, OpenStack was added to these by Piston Labs.
Christof did manage the impressive task of explaining the architecture of cloud foundry in the dark as the automatic lighting and power controls for the conference room turned off in the middle of his talk. When the power returned, Marti took us through the different deployment models for 1 to 5,000 VMs.
We’re reserving the openstack.ch domain for the Swiss user group for further communications and plan for the next meeting in around two months. For those who want to discuss before then, the mailing list at http://sympa.systemsx.ch/sympa/subscribe/swiss-openstack-usergroup is available.
It was great to see the enthusiasm continue amongst the attendees and many thanks to SWITCH for hosting, Rackspace for the pizza and ICCLab for the beer.
For those interested in joining the OpenStack User Group channels.
General information about the group