Vancouver Systems

We are excited to have two fantastic talks lined up.

The digital systems we build shape the society we live in. Despite this, as views on its direction change rapidly, the architecture of our distributed systems (thin clients on consumption-oriented devices offloading all compute to foreign hyperscalers) hasn't changed, both for a genuinely lack of replacement tech and gatekeeping by Apple and Google, making them inaccessible to most. Thanks to very recent P2P improvements, new personal computing platforms built entirely on Free Software and Open Hardware, and regulations finally factoring in the risks technologists have advocated for decades, this is changing rapidly, forcing us to fundamentally re-think how apps are designed. This talk introduces a new stack to fix this, from the lowest level, hardware (RISC-V) up through the kernel (Linux), userspace (systemd/GNOME/Flatpak), app design (libadwaita), new local-first, post-internet transports (Bluetooth, NAN, Mesh, sneakernet) and the high-level libraries for distributed systems on top of them (p2panda, Loro CRDT), with the goal of making this new generation of apps just as user-friendly as the existing ones, if not more.

Felicitas Pojtinger (@pojntfx) is Head of Research and Development at Loophole Labs, where she works mostly on low-level userspace and live migration of containers and virtual machines. By night, she contributes to various Free Software projects as a member of the GNOME Foundation and a board member at VanLUG. She's the developer of weron, a WebRTC overlay networking tool, bofied, a Go network boot server, the go-nbd library, and more.

Building a Lightweight Data Streaming Framework on Apache DataFusion

Apache DataFusion is a very popular extensible query engine written in Rust that uses Apache Arrow as its in-memory format. Many analytical databases and query engines use DataFusion as a foundation. But can you extend it to build a data streaming framework? In this talk, I'll cover the lessons learned from extending DataFusion to support streaming. We'll look into how well most of the existing components can be leveraged with little or no modifications. We'll also see how modern features, like implementing TypeScript support via WebAssembly, can be integrated with the Arrow format.

Yaroslav Tkachenko is a Software Engineer, Consultant, and Advisor specializing in data streaming & data-intensive applications. Currently, Yaroslav is a Founder at Irontools, consulting companies that need to design, build and evolve efficient data streaming systems. Previously, Yaroslav was a tech lead at Shopify, Activision, and several startups.

Many databases are pivoting to how to address scalability as workloads are dominated by AI agents rather than human developers or ORM generated applications. And while distributed SQL databases such as TiDB handle the scale of the system, the reliability of the application becomes equally as critical. This session delves into why query optimizers struggle to deliver reliable SQL performance, and how TiDB is solving this to be more adaptable given less predictable workloads.

Terry Purcell is the Chief Optimizer Architect at PingCAP, leading the development of the TiDB query optimizer. He has over two decades of experience in query optimization, including 10 years leading one of the most reliable cost-based optimizers in the industry. Terry has contributed to more than 40 US patents, as well as countless webinars, white papers, articles, and conference presentations.

CHERI is a set of architectural concepts and mechanisms that add capabilities as a fundamental primitive for memory access. You may have heard it described as "hardware-backed memory safety." But neither description is all that satisfying. Does CHERI make my C code memory safe? What is memory safety, exactly? And if my codebase is already written in a safe language, why should I care?

Milo Piccini Noble is a 4(+1)th year undergraduate student at UBC, and a contributor to the CHERIoT project, where he's extending the shared heap allocator. Milo's interests span security, languages, systems, and developer tools: really, how we can build software that doesn't break in terrible ways. He has a lot to learn.

CUDA over TCP - how we reverse engineered the CUDA API to allow us to run CUDA workloads in environments where there is no physical GPU. A live demo would include running a large open source LLM model on a MacBook!

Shivansh is a long-time open-source developer who started his infrastructure journey back in 2018 at IBM where his work centered around various container and Kubernetes platforms. He was also actively involved in the early days of service meshes, and today his goal with Loophole Labs is to finally fix the core infrastructure problems that plague the industry.

This event is sponsored by TiDB and Antithesis.