Linaro Virtual Connect 2020
In early 2020 we were looking forward to our twice annual Linaro Connect event and to welcoming the 400+ Linaro employees, assignees, members, partners and others in the open source community. Well, as you all know things in 2020 haven’t quite gone as planned but we are happy to be able to host Linaro Virtual Connect as a way to preserve some of the things we love most about Linaro Connect.
The shift to a virtual event has been a learning experience for everyone - events team, speakers, and attendees and we thank you for your patience and willingness to try something new. One benefit of a virtual event is that attendees and speakers who have not been able to attend Linaro Connect in the past have had the chance to participate.
We are very happy to have so many of you invested in Linaro Connect and pleased that we are still able to deliver the important technical content.
With all the changes 2020 has brought, Linaro Connect continued to the tradition of quality sessions at the forefront of Arm ecosystem development. The three days were staggered at different start times to accommodate as many time zones as feasible. With a large percentage of sessions prerecorded, we were able to feature speakers from around the world.
There were an astounding 1,476 individual registrations, three times the in person Connect average. Live conversations started during the sessions and continued in the Slack channels afterwards. For a change of pace, Kassidy Holmes led participants through a 1 hour foundational yoga flow on Wednesday. On Thursday, Martin Jackson wrapped up the week with a lively acoustic set.
The most popular sessions (by registration counts) were.
- Arm64 Linux Kernel Architecture Update
- Trusted Firmware Project Update
- Arm Architecture 2020 Extensions
- PSA Secure Partitions in OP-TEE
- Enable UEFI Secure Boot Using OP-TEE as a Secure Partition
Tuesday September 22 Highlights
Li noted that Linaro ranked #5 in patches committed in the Linux kernel between 2007-2019, even though Linaro is only 10 years old. Li noted that new companies are starting up at a brisk pace. Companies are making their own chips, carrying differentiation from the system level to the chip level. The cost of chip design has significantly declined making custom chip design more affordable. This is a way to secure access to IP and supplies particularly in the face of on-going shortages and trade wars (see China/US). This of course has a direct impact on SoC vendors. In Li’s view, the Arm ecosystem is composed not only of hardware vendors but now major software vendors are playing a role. Linaro is enlarging its scope for these new companies and the software vendors with the idea of a franchise company. A franchise, in this context, is a sub ecosystem within the Arm ecosystem. An example is Google and Android. Linaro also has simplified its membership model to four tiers - Core, Club, Group and Project, while continuing to focus on delivering value through its corps of maintainers and skilled developers. Finally, Linaro is opening up its projects and processing to a worldwide audience.
While “Edge” means many things to many people, in order for Edge solutions to be a success, the key is using the Open Standards Data Center model. Edge platforms face several challenges including environmental, scale and cost. Which will vary depending on which Edge “tier” that platform plays in. Open standards, as with the IBM PC and various networking standards (TCP, HTTP), lower barriers to entry and protect investments in time, money and knowledge. Enterprise standards are useful. In addition to knowledge reuse, you are not locked into a single platform. You can use the right hardware with the same base software stack using the same or similar security models and processes. Provisioning and on-boarding end devices need to be deployed with general knowledge at large scale (10 of thousands) across geographic regions. Using OCI container solutions provides consistency and scalability. Standardisation in edge computing giving manufacturers and consumers more, not less, choice.
As usual Grant gave a practical talk on direction and progress in the complex space of edge and IoT device booting. Platforms are only created when interfaces are stable and reliable - there is less work per platform enablement
- Stage 0 : Enable UEFI on U-Boot
- Stage 1: Rune UEFI Self Certification test (SCT)
- Stage 2: Interpret SCT results; currently lacking a parsing tool to interpret results
- Stage 3: Boot an unmodified OS to a functional state; Device Tree is critical here but support is inconsistent
- Stage 4: Test multiple deployment scenarios; work required here; can use some tools from the Data Center world.
Matteo Carlini and Shebu Varghese Kuriakose
The Trusted Firmware (TF) project’s mission is to collaboratively build a secure reference software implementation for Arm processors. Highlights from the past year include the addition of Mbed TLS (donated by Arm) and Hafnium (donated by Google) to TF. TF for Cortex-A (TF-A) v2.3 and TF- Cortex-M (TF-M) v1.1 were both released. Renesas and NXP joined as members. Also a new Security center was set up to provide consistency on handing security vulnerabilities and incidents. A new maintainer process was put in place including how code reviews and the patch lifecycle are managed. Half of the TF maintainers are now from outside of Arm. The project also announced that Don Harbin (Linaro) will be the TF Community manager and that the TF website has been significantly updated. Looking ahead, TF is expanding CI/Testing efforts with more platforms, static analysis and updated user guides. The community project will also be sponsoring workshops on TF-M and Mbed TLS.
A don’t miss session is LVC20-104 On the Edge of the Real World. An Introduction by Bruno Verachten. A lively talk from Bruno on how to build home IoT systems. The bottom line? There’s a lot of choice!
Zhenyu Zheng and Kevin Zhao
Currently, one big gap between Arm64 and X86 cloud platforms is that X86 can provide a much better instance migration experience than the Arm64 platform. CPU comparison and CPU model capabilities have provided the Arm64 VM with the ability to live migration among different hardware vendors. This function is the essential function of the data center. From the cloud management framework, we also need to consider the realization of supporting VM live migration. In this session, we discussed what we have done in the most widely used virtualization management tool - Libvirt to provide better live migration capabilities on Arm64 platform and also some details in the newest lightweight cloud management project such as Kubevirt. With live migration support on Arm64, it can finally benefit the cloud ecosystem for large scale datacenter scenarios which may use different Arm64 CPU architectures and vendors. In the process of the work the team discovered missing or lack of functional code for Arm64. It is possible these issues exist in other code areas.
Wednesday September 23 Highlights
Jean Marc Denis
The goal of the European Processor Initiative (EPI) is to become one of the leaders in High Performance Computing (HPC). EPI is a European backed initiative to develop a complete European designed high end microprocessor addressing the Super Computer and edge-HPC segments. This week, the EU committed €8 billion for this effort. Europe wants sovereign access to high performance, low power microprocessors from initial IP to delivered products and reduce the dependence on non-EU suppliers. SiPearl is the industrial partner of EPI operating as a commercial entity to benefit the EPI project and its members. The plan is to have a unified environment based on Arm Zeus cores by 2023. Different classes of accelerators will be used as part of the workflow engines. The main strength of Arm is IP going from device to the super computer. The presenter’s view is that the NVidia/Arm alliance is a unique, fantastic combination of AI and GPU capabilities.
SiPearl’s first target market for Rhea chips is HPC. Then in order Cloud, Edge (which is similar to Cloud requirements) then Automotive/Industrial Edge. Arm is attractive as it has a fully fledged ecosystem. Good hardware needs good software. The Arm ecosystem is self-sustaining. “Arm wins because it has an ecosystem”. Rhea is a hyperscale processor using the Arm Zeus core with coherent on chip network; There is a need to scale from chip to HPC to Edge applications. It is important to include high bandwidth memory with low power, low latency links using state of the art design and architecture. The plan is to make the EPI Common platform open standard in the future. Rhea is intended to be a General Purpose Processor balanced between performance and memory bandwidth. You will also need accelerators to provide optimization.
Carlo muses on the cultural basis for license compliance in the Open Source community. The pillar is respect, not fear of litigation or fines. Enforcement does not happen that often and companies are getting better at compliance. Carlo posits that compliance comes from social norms - the right thing to do. If rules are simple and straightforward as well as reasonable and self-evident, compliance will be nearly uniform. The standard Open Source licenses are reasonable and straightforward so it is easy to comply. In order to receive respect, you must give respect. People can pledge to a higher standard using such as Open Chain and SPDX in headers. One way to show respect is to take a bit of effort to ensure users of your code can easily find the licensing and restrictions in your code.
Poonam Aggrwal, Priyanka Jain and Ilias Apalodimas
The Embedded Base Boot Requirements (EBBR) specification defines requirements for embedded systems to enable interoperability between SoCs, hardware platforms, firmware implementations, and operating system distributions. EBBR is targeted at making operating system/distros agnostic to the platform. Same operating system image should run on any hardware with a well-defined firmware interface which is EBBR compliant.
There has been significant work going on in U-boot with regards to EBBR in the open-source community. Various features like bootefi are already available with many other features in queue.
This presentation aims at explaining how EBBR specifications gets mapped to NXP platforms and demonstrating EBBR compliance for NXP platforms. The reference boot-architecture will be based on TFA, u-boot, device-trees, Linux and OPTEE (for secure uefi flow). This will demonstrate distros like SUSE running on NXP SoCs using bootefi command, secure uefi flow, etc
Also efforts are going to ensure that the U-boot is EBBR compliant by running FWTS, SCT for EBBR. The idea is to make the u-boot feature complete and can be demonstrated as EBBR compliance on NXP devices.
Mike Holmes, Azzedine Touzni and Alex Bennée
Virtualization fireside chat moderated by Mike Holmes Director of Engineering Foundational Technologies at Linaro featuring guests: Azzedine Touzni Sr. Director of Engineering at Qualcomm, Alex Bennée Sr. Software engineer at Linaro.
This fireside chat covered topics such as the current challenges in the virtualization domain, the state of virtualization on Arm, the end vision for Stratos, how to get involved, and more…
The scheduler has been the place of a lot of changes during the past releases with new interfaces to set properties of tasks and/or groups of tasks; Other evolutions are ongoing and this session will go through the main changes merged during the past releases and the ongoing discussions for next changes.
Thursday September 24 Highlights
Jose Manrique López de la Fuente
Jose started quoting William Edwards Deming “Without data, you are just a person with an opinion”. So what to do with data - “Strategy without tactics is the slowest route to victory. Tactics without strategy is the noise before defeat.” - Sun Tzu
Data must be matched with context knowledge. Looking at the Linaro Use Case, there is a huge amount of contributors affiliated to “Unknown” organizations. Why? Is data wrong?
Jose notes that every company is becoming a software company. And how a company interacts with the software ecosystem can help or hurt them. Success is understanding the (1) Legal aspects (2) People and (3) Engineering and the give and take between them. Jose recommends that you check out howtogroup.org and chaoss.community for wisdom. The best approach is to combine strategy, tactics and context knowledge to approach open source and generated analytics.
John Stultz & Todd Kjos
This session was an Interview (by John Stultz) with Todd Kjos, Google, about the reason for, and challenges with, Generic Kernel Image (GKI) project. Fragmentation is in the way of keeping Android devices up-to-date with latest kernel patches (bugs,security, h/w, etc) - based on LTS kernel. Google is keeping Pixel devices up-to-date but few others are doing the same. A module kernel approach to allow Google to keep the kernel up-to-date and allow vendors to maintain their customizations. Tools keep track of ABI stable symbols (KMI - Kernel Module Interface - subset of the full kernel ABI). Google strongly recommends vendors to send their changes into the upstream kernel. Benefit of common use and less fragmentation (when differentiation isn’t actually required but has happened in the past). Therefore helping vendors to eliminate unnecessary technical debt.
Tomas Evenson & Nathalie Chan King Choy
The OpenAMP Linaro Community Project is focusing on standardizing aspects of embedded heterogeneous software through open source projects. Current SoCs are very heterogeneous. In the past, integrations of the various computing components has been ad-hoc. The goal of OpenAMP is to standardize the interactions between the components. Project has a Board, a Technical Steering Committee and working groups ( System Device Tree, Remoteproc / RPmsag, Virtio and libmetal)
Anders Roxell and Paolo Valente
LKFT has been endowed with I/O benchmarks. They measure both I/O throughput with general workloads, and system-and application-level latency under heavy loads. More benchmarks are being evaluated as well. In addition, the current I/O benchmarks are being used as a pilot case, to make a general solution for automatic detection of performance regressions. In this presentation we will describe these interesting developments.
Kumar Gala & Paul Sokolovsky
Testing MCU’s (Cortex-M) systems presents challenges different from Cortex-A/Linux systems. Using LAVA infrastructure and tools, the team has worked with the LAVA and LAB teams to evolve the apps and tools to support continuously running tests in a CI context. A key piece was to utilize containers to provide consistency and the ability to run in multiple environments. Kumar Gala walked through a short demo on what this looks like a developer system. Paul Sokolovsky spoke about using this setup for network testing with the result of finding serious regressions prior to the October release of Zephyr.
It seems the Rust programming language comes up frequently these days, and there is a lot of interest in it. We have been using Rust to implement a simulation test environment within MCUboot since 2017. This presentation will discuss our experience with Rust, and our hopes of how this language might help other projects in the future, especially in regards to security.
Big Data is a huge ecosystem of a large number of applications and companies. Linaro’s goal is to ensure that Arm platforms are on par or better with other architectures in as many areas as practical. As part of this effort, the team is benchmarking against x86 and optimizing to leverage AArch64 advantages. Achieving this influence has taken time and effort to encourage projects and applications to consider the Arm server platform. The focus has been on Apache Big Top as it is a foundation for many other applications and products. LDCG has been a leading contributor. The next release of Big Top is v1.5 in a few weeks. The team is also looking at Apache Ambari - a management and security application.
This presentation presented a real world challenge - how to analyze the results from Full Wave Inversion (FWI) testing. This is used in oil and gas exploration, active drilling and reservoir prediction. This is a convergence of HPC and AI as the proposal is to use Deep Learning to build models. The speaker notes that this work is highly parallel processing. As an example, one company put together 250 Petaflops with 40,000 Intel Xeon Phi 7250 (68 cores per socket) to tackle this problem. To date, the high computing cost has discouraged work in this area. The speaker is proposing an academic- Industry alliance to tackle this problem through the HPC-AI project.
With distinguished presenters from around the world and from some world class leading businesses, our Virtual Connect delivered 65 sessions. This content covers some of the groundbreaking topics of the moment and will be added to our already vast amount of resources that has been compiled over the last 10 years ago, when Linaro was first founded.
Although the current climate has curbed our celebrations for our anniversary, it has not curbed our achievements and we are proud of our successes in working together with our members and the community alike.
All videos and slides from presentations are now available on our Resource page: https://connect.linaro.org/resources/lvc20/
We look forward to our next in person event when we can enjoy some of our favorite aspects of Linaro Connect- team hacking time, Joe Bates’ morning fun facts, the “Ask Arm Anything” session, socializing after hours with colleagues and friends, Demo Friday, and everyone’s favorite: Dave Pigott’s puzzle.
Thank you for attending Linaro Virtual Connect 2020.
Until next time!
Authors: Vicky Janicki Kristine Dill, Jonathan Burcham, Mark Orvek and David Rusling