The OpenStack Foundation held their last biannual OpenStack Summit in Tokyo, Japan, from October 27-30, 2015. OpenStack Summits include keynotes, in-depth technical discussions, hands-on workshops, and the full presence of almost every key player in the OpenStack ecosystem.
I attended the summit together with three other colleagues from it-novum, from Tuesday till Friday. In this article, I'll try to summarize my impressions and share my notes and observations.
The Tokyo summit took place in the Grand Prince International Convention Center & Hotels, which was easy to reach via public transport. The conference venue was spread across several buildings and floors, which sometimes made it difficult to navigate and find the right room. Fortunately, friendly helpers were present everywhere, to provide guidance and directions. Overall, the conference was organized in a very professional manner and it was a pleasure to attend.
My focus was on attending storage-related sessions as well as presentations about Linux Containers (e.g. Docker) and other OpenStack-related tools for automation and deployment. I attended mostly sessions from the main conference, but also jumped over to some sessions in the Design Summit from time to time, which was held in parallel to the main conference in a separate building. The schedule was quite intense - I quickly realized that there were sometimes 3-4 sessions taking place in parallel that I would have loved to attend!
Between the larger breaks, there were only 5 minute breaks between the session slots, which sometimes made it difficult to reach the next session in time. Moreover, some rooms were fairly small and more than once the rooms were overflowing by the time I reached the entrance. Fortunately, most of the sessions were recorded on video and were made available within hours after the presentation was held . The availability of video recordings somewhat eased the pain of not being able to attend multiple sessions of interest that took place at the same time.
In addition to the ongoing sessions, there was the "OpenStack Marketplace", a large exhibition floor located in the basement of a building next to the convention center. The marketplace was somewhat remote, but all of the key players/vendors in the OpenStack ecosystem (SUSE, Canonical and Red Hat, Mirantis and hardware vendors of all brands) were present. I personally enjoyed meeting the founders of Quobyte from Berlin, to learn more about their interesting distributed storage system, which is based on their former research project "XtreemFS".
To keep track of the schedule, the mobile App was invaluable. However its usability and performance left a lot to be desired: it did not provide an offline mode and did not take the current time and date into consideration, which resulted in a lot of unneccesary scrolling.
The social event on Wednesday night also was a memorable event. This year, the key sponsors collaborated on organizing one big party instead of organizing several separate ones. The event took place in the "Happo-En", a beautiful Japanese garden with a pond. In addition to a wide variety of food, it provided lots of activities and performances: sumo wrestling, geisha performances, Taiko drum lessons, sake tasting, Cosplay...
The entire event had a strong focus on community and collaboration. The term "We are OpenStack" and the #weareopenstack hash tag were promoted throughout the conference, to share moments and highlights with other participants.
Another thing that impressed me was the availability of simultaneous translations for sessions held in Japanese. Attendees could obtain headphones when entering the room, to listen to these presentations in English.
Notes and takeaways
Tuesday and Wednesday started with keynotes, that were also broadcast in additional "overflow rooms", to accommodate the large number of attendees that could not fit into the main auditorium. Some personal highlights from the various keynotes I attended include:
- The OpenStack Project Navigator, to help maintaining an overview about the adoption, age and maturity of the various OpenStack subprojects. This will help new users to obtain some guidance on what projects are safe to use, and which should be used with caution.
- The fact that Software defined Networking (SDN) grows twice as fast as server virtualization.
- Erica Brescia's (COO BitNami) keynote Banishing the Shadow Cloud. This was a thought-provoking and insightful presentation, while most other sponsored keynotes were somewhat too product centric for my taste.
Below is a (likely incomplete) list of sessions I attended, including some notes and key takeaways. I've also taken some pictures during the Summit, which can be found in my OpenStack Summit Flickr Set.
I also managed to sneak in some time for sightseeing, you can take a look at my pictures of Tokio here.
This session was a very good introduction into the topic of "object storage" in general, giving an overview about the general concepts and access methods and then covering OpenStack Swift in more detail from a developer's perspective. Christian then built a simple web application that used Swift to store the actual data to provide a real-world use case, followed by some "Do's and Don'ts" best practice advice.
In this fast-paced session, Mike gave a speed-dating type overview about various Linux container technologies, related tools and frameworks that help to manage and orchestrate large-scale container deployments. It covered a lot of technologies just briefly, so this was a useful session if one had some general knowledge of containers already and just wanted to get a broader overview of the container ecosystem. However, there was no real conclusion except there is a lot of choice, so choosing the most appropriate tool for the job at hand is still up to the user. This session gave me a lot of pointers and hints about what tools might be worth evaluating, but it also left me somewhat puzzled and overwhelmed by the sheer amount of choices that were presented.
Life Without DevStack: Upstream Development With OSAD by Miguel Grinberg (Video: https://youtu.be/OSkGaxR1yds)
In this talk, Miguel talked about the pains and problems that he was facing when using DevStack for performing local development on certain OpenStack projects. The proposed alternative "OpenStack Ansible Deployment (OSAD) distribution" (especially the bootstrap-aio.sh script) looked quite promising: instead of installing all OpenStack components on a sinlge node, this script puts every OpenStack component in a dedicated LXC container, using Ansible for the assembly and configuration. If you're an OpenStack developer, this tool might come in very handy. He also covered this project in a detailed blog post.
Yuting Woo from AWCloud summarized his findings comparing Ceph with proprietary solutions from Solidfire and ScaleIO in terms of storage functionality, deployment options, operations and maintenance as well as performance. Based on his findings, there is still a lot of room for improvement in Ceph, particularly in the areas of deployment, management, performance as well as advanced storage features like QoS (Quality of Service), Deduplication or Compression.
The Cinder core developers shared their view of the state of Cinder and where the project is heading. API versioning was mentioned as an issue, and API microversioning was discussed as a solution. Other tasks under development include: performing backups of snapshots instead of volumes, as well as implementing TRIM support. The Cinder developers are also contemplating to decouple Cinder from OpenStack, to convert it into a standalone, general purpose service for other cloud technologies other than OpenStack.
In this session, Jesse argued if having multiple distributions (or "flavors") of OpenStack is really helping the project or actually harming it. He also asked what OpenStack users really want in order to get started with it. He observed a trend that many users would like to start using OpenStack, without actually having to setup and deploy the infrastructure by themselves. He sees this as a market opportunity for companies, to set up and manage OpenStack infrastructures on behalf of their clients, in their own data centers.
Walk Through a Software Defined Everything PoC by Chris Janiszewski, Sandro Mathys
This presentation talked about a proof-of-concept (POC) installation of OpenStack, using Ceph as the storage backend and MidoNet as the network virtualization layer used by Neutron. The speakers shared their experiences and key learnings from this project. One interesting insight was their choice of using xCat (Extreme Cluster/Cloud Administration Toolkit) for the provisioning and management of the OpenStack nodes themselves.
Ceph Community Talk on High-Performance Solid State Ceph by Reddy Chagam, Gunna Marripudi, Allen Samuels, Warren Wang
In this session, representatives from various storage companies (Intel, Samsung, SanDisk) talked about the projects their are working on to improve Ceph's performance on solid state devices. A lot of effort is being put into creating benchmarks and other profiling and testing tools. For example, Intel has worked with Red Hat to release the Ceph Benchmarking Toolkit (CBT) as open source. SanDisk is also putting a lot of effort into improving Ceph's performance, by submitting patches that make the entire code base more scalable and to resolve some of the bottlenecks they observed. In general, some of the assumptions about spinning disks that influenced design decisions in Ceph are now being put on the table for re-examination and possible ways of improvement. This talk was very informative and reinforced the impression, that Ceph's existing performance deficiencies are goint to be adressed over time.
New Ceph Configurations - High Performance Without High Costs by Allen Samuels
Allen Samuels presented some recent performance improvements and shared some benchmark results that SanDisk has been working on to improve Ceph's performance. He also compared the use of erasure coding versus traditional replication and cache/tiered storage pools in combination with flash devices.
Ansible Collaboration Day: Ansible + OpenStack — State of the Universe by Robyn Bergeron, Mark McLoughlin
This was an open meetup of Ansible users and representatives from Red Hat (who acquired Ansible a few weeks before the OpenStack Summit). It included a presentation of upcoming features in Ansible 2.0, which is highly anticipated by the community and will include a number of useful improvements. The release seems to be imminent, but no firm date was given. For the Ansible team, not much semms to have changed after the acquisition. The question if Ansible Tower will be released under an open source license was raised, and the expectation is that the Ansible acquisition will be no differerent than other acquisitions in terms of what happens to previously proprietary components (e.g. Ceph's Calamari, which was released as open source after Red Hat acquired InkTank).
Turning Pets Into Cattle: A Demonstration to Provoke Discussion by Yih Leong Sun, Stephen Walli (Video: https://youtu.be/wL9FOHKpmCU)
In this session, Stephen and Leong discussed some of the aspects of moving existing applications from dedicated servers into a cloud enviroment. Even though this session was intended to be "non-technical", it contained a live demo in which the web tier of a WordPress application was migrated into the cloud, while keeping the database service on premise. In a following step, static content was pushed into a Swift object store. I found the mix of high-level business aspects with a live shell demo somewhat confusing, even though the session itself contained a few insightful recommendations and insights.
Kolla: Ansible Deployment + OpenStack in Docker Containers = Operator Bliss Steven Dake, Daneyon Hansen, Sam Yaple (Video: https://youtu.be/BKYJuYsT4z4)
Steven and Sam introduced Kolla, an OpenStack deployment tool based on Ansible and Docker that supports a wide range of Linux distributions (RHEL and derivatives like CentOS or Oracle Linux, Ubuntu Linux). The session included a live demo to explain Kolla's functionality, that deployed a functional OpenStack environment (including Ceph) in about 20 minutes. Kolla is developed by a diverse community, including engineers from Red Hat, Cisco Systems, Oracle and Rack Space. It was interesting to learn that this is not only used for setting up OpenStack development environments, but also for real-life production environments as well. The development is guided and steered by the community and is designed for scale and openness, including a thorough review and testing process.
Persisting Data In Your Cloud With Cinder Block Storage by John Griffith, Kenneth Hui, Arun Sriraman (Video: https://youtu.be/qeFz0pwVO6c)
This session started as a basic introduction into Cinder and the benefits of using OpenStack as a developer, but quickly turned into a product showcase for Platform9 and SolidFire. For someone new to Cinder, the first part of the session was quite useful to get up to speed, the second part showed a real-life production configuration that could be used as a blueprint for own deployments.
Finally FDE - OpenStack Full Disk Encryption and Missing Pieces by Robert Clark, Dave McCowan, Arvind Tiwari (Video: https://youtu.be/phMUoq9OYHY)
One of the key issues with encrypted disk drives is the management of the encryption keys. In an OpenStack environment, it is very impractical to wait for the administrator to log into the console and enter the decryption passphrase every time a virtual machine reboots. In this session, Robert, Arvind and Dave gave an overview about the state of encryption in various OpenStack projects and also introduced Project Marshal and Project Leeson. The latter implements a key broker service based on the UUID of the guest OS. Securing data in a cloud environment is a big challenge, this session provided a good overview and some practical examples on how data could be encrypted without too much manual effort.
Ceph and OpenStack: Current Integration and Roadmap by Josh Durgin, Sébastien Han (Video: https://youtu.be/5Wqb5sZG11I)
Josh and Sébastien from Red Hat gave an overview and status update on Ceph and OpenStack support. They started by giving a general overview about Ceph and its components, followed by an introduction of how Ceph integrates with OpenStack, e.g. as a backend for Keystone (using the Swift API support provided by RADOS Gateway), Cinder, Glance and Nova. A lot of bugs in the interoperabilty with Ceph were fixed in the OpenStack "Liberty" release. New features like Cinder volume migration were also introduced in this release cycle.
The Ceph "Infernalis" release is also close to being released, which will include new features like per-image metadata, flattening of snapshots and easier deletion of parent images. The "Jewel" Ceph release will also include many highly desired features. For example, CephFS might actually be declared stable in Q1 2016 - according to the developers, the lack of a functional file system check tool (fsck) is one of the main remaining road blocks. Write performance improvements are also much appreciated.
This session was packed with news and interesting developments. If you're interested in where Ceph and OpenStack are heading, this session was a real gem.
Beginners Guide to Containers Technology and How it Actually Works James Bottomley (Video: https://youtu.be/YsYzMPptB-k)
This was the most thorough and in-depth presentation about Linux containers I have ever attended, going down to the actual Linux kernel APIs that provide the foundation for Linux containers (regardless of their actual implementation, e.g. Docker, LXC or LXD). After explaining the general benefits of using containers over virtual machines (VMs), James did an excellent job of explaining and demonstrating the various low-level systems, e.g. name spaces or control groups. Unfortunately he ran out of time and had to rush through the rest of the presentation. Also worth mentioning was that the presentation was created using the Impress.JS presentation framework, which provided a very welcome change to the many "death by PowerPoint" presentations.
This was a joint presentation by representatives from NetApp, Red Hat and SUSE about the OpenStack Shared File Service "Manila". It started by giving an overview of Manila, its use cases and what's new in the OpenStack "Liberty" release. The representatives from SUSE and Red Hat then talked about the status of manila on their distributions. SUSE aims to fully support Manila with SUSE Cloud 6, Red Hat is working on CephFS support (using NFS-Ganesha to export file systems via NFS). They concluded the presentation with an outlook of what to expect from the upcoming "Mitaka" release and a live demonstration of the "share replication" feature, which allows creating high available NFS shares (if the backend supports it).
To summarize my personal impressions and key learnings as a first-time attendee: OpenStack Summit is huge! Don't even try to attend all the sessions that interest you, otherwise it will be a frustrating experience. Plan your sessions in advance, and also plan to allocate some time after the event, to watch the recordings of sessions you were not able to attend. Also, take notes! Otherwise the sheer amount of information you have to digest in such a short period of time will overwhelm and confuse you.
For more hints, the session Proud to be a Noob: How to Make the Most of Your First OpenStack Summit is probably a useful resource. Unfortunately I missed this one...
In addition to learning a lot about recent developments and ongoing activities in the OpenStack ecosystem, the networking and meeting people behind the projects was probably the most valuable benefit for me. This kind of interaction was only possible by attending the Summit in person, instead of watching the video recordings offline. I look forward to attending the next!