Engineering at Meta

The Jump-Start technique has significantly improved the warm-up and steady-state performance of the HipHop Virtual Machine for Facebook.com. These improvements make the site faster to load and allow us to roll out new features much more quickly. https://engineering.fb.com//developer-too/hhvm-jump-start/

Undetected data errors within hardware, a widespread problem for large-scale infrastructure systems, can lead to data loss and take months to resolve. In new research, we provide strategies to detect these types of errors and counter their impact at scale.

Here’s how we designed the Facebook Ordered Queueing Service (FOQS). It supports the asynchronous computing workloads that run many of Facebook’s services at scale. https://engineering.fb.com//foqs-scaling-a-distributed-pr/

Our engineers commit thousands of diffs into production every day, it’s hard to know which to prioritize first. Earlier detection is the top way to reduce fix times. We’ve implemented efforts to find the right engineer quickly, get fixes done earlier in the development process, and reduce alerts, creating a more effective way to help engineers easily see what to work on first, and spend less time debugging and fixing issues. Learn more: https://engineering.fb.com//02/17/developer-tools/fix-fast/

Root cause analysis is a key part of fixing any code bugs. We’ve developed Minesweeper, a technique used as a first line of defense at Facebook to identify bugs and automate fixes based on their symptoms, helping prevent potentially wide-scale disruptions. More: http://ow.ly/6P9h50DvStU

The Facebook News Feed presents personalized content to each of the more than 2 billion people who use it. We are sharing new details of how we designed an ML-powered News Feed ranking system and the layers of ranking that take place in real time to serve up the content you see in your feed. https://engineering.fb.com//ml-applica/news-feed-ranking/

2020, what a year it’s been! As stay-at-home orders put a strain on infrastructure across the globe, we focused on keeping our services reliable, expanding access to connectivity, and using renewable energy to power our operations. Here’s a look back: https://engineering.fb.com//connectiv/2020-year-in-review/

We built rsys a smaller, faster video calling library to streamline video calling across all our apps, with multiplatform support. With this new iteration, we’re reimagining how we think about our video calling platform and starting from the ground up with a new client core and extensibility framework. This helps us advance state-of-the-art technologies with a new codebase that’s designed to be sustainable and scalable for the next decade, laying the foundation for remote presence and interoperability across apps. https://engineering.fb.com/2020//21/video-engineering/rsys/

The Evenstar program is testing the first interoperable radio access network component for 4G/5G networks. OpenRAN adoption helps improve infrastructure for global networks, accelerate adoption, and strengthen connectivity in rural communities while lowering costs for mobile operators. https://engineering.fb.com/2020/12/17/connectivity/evenstar/

When the pandemic began, we turned our focus to new ways to help people stay connected and hang out together now that we’re all apart. To ensure that people see and hear the same thing at the same time in these new shared synchronous experiences, like Rooms and Watch Together, we needed to build a brand-new state synchronization platform over our video calling stack. https://engineering.fb.com/2020//16/android/co-experiences/

Rajeev Rajan, our Director of Engineering for Video, will be giving the keynote at tomorrow's virtual Video@Scale event. See him and others talking about managing video infra at scale. https://atscaleconference.com/events/videoscale-remote/

We are replacing the de facto protocol the internet has used for decades. The majority of our internet traffic is now using QUIC and HTTP/3, providing a significantly improved experience for people on Facebook and Instagram. https://engineering.fb.com//how-facebook-is-bringing-quic/

Facebook AI is working with Carnegie Mellon to find new ways to store and use renewable energy -- and releasing a new data set, baseline models, and benchmarks to accelerate progress in this important area.

resctl-demo is an open-source simulation and testing tool that helps developers resolve system resource conflicts. https://engineering.fb.com//introducing-resctl-demo-bette/

To help us quickly find the most current information to make informed decisions, we’ve built Nemo, an internal data discovery engine. In day-to-day use, Nemo has improved search success rates, simplified the data discovery process across different engineering teams, increased data usage quality, and reduced overall time spent looking for the right data. https://engineering.fb.com/data-infrastructure/nemo/

We’ve open-sourced CG/SQL, a code generation tool for the popular SQLite library. CG/SQL makes it easier to create accurate code for even highly complex stored procedures with very large queries.

Facebook CTO Mike Schroepfer recently shared his thoughts on how a transparent culture helps engineers do their best work: https://www.facebook.com/careers/life/unlocking-creativity

Congratulations to Facebook London’s Mark Harman who has today been elected as a Fellow of the Royal Academy of Engineering Fellowship for his software engineering research and work on Sapienz and Web Enabled Simulation. Royal Academy of Engineering

Facebook AI is releasing KILT, a unified benchmark to help AI researchers evaluate #NLP models across multiple knowledge-intensive language tasks.

We believe the next 10 years will be the defining decade in the world’s efforts to dramatically reduce GHG emissions, limit global warming, and mitigate the impacts of climate change. This is why we are committed to tackling climate change through our global operations, value chain, and beyond. Today, we are sharing our commitment to reach net zero emissions for our value chain in 2030. https://engineering.fb.com/data-center-eng/net-zero-carbon/

We use throughput autoscaling, which estimates and predicts capacity requirements, to manage the load on our services and balance demands across all the machines in our data centers. Learn more: https://engineering.fb.com/networki/throughput-autoscaling/

How optimal workload placement allows Facebook’s systems to tolerate sub-data center faults without redirecting traffic away from the entire data center. https://engineering.fb.com//fault-tolerance-through-optim/