Over the past few years, we've been working to upgrade our data centers to run at 100 gigabits per second. To do so, we needed to deploy 100G optical connections to connect the switch fabric at higher data rates and allow for future upgradability — all while keeping power consumption low and increasing efficiency. We created a 100G single-mode optical transceiver solution, which we've shared through the Open Compute Project.
פוסטים
Bryce Canyon, our next-generation high-density storage server, is designed to support more powerful processors and more memory, and improves thermal and power efficiency by taking in air underneath the chassis. Our goal was to build a platform that would not only meet our storage needs today, but also scale to accommodate new modules for future growth.
סרטונים
פוסטים
Dive deeper into Big Basin, our new GPU server that can train machine learning models that are 30 percent larger.
Today at the Open Compute Project Summit we announced an end-to-end refresh of our server fleet: Big Basin, our new GPU server; Bryce Canyon, our high-density storage server; and updated compute platforms in Tioga Pass and Yosemite v2. Read more about the hardware specs and performance improvements here.
At today's Video @Scale event in Menlo Park, Facebook announced updates to its dynamic streaming technology to deliver high-quality 360 videos more efficiently.
Over the years, as our community and data needs have grown, Facebook's data infrastructure team has explored new solutions for processing data at scale. They decided to use Apache Spark for its ability to scale with a large amount of data and support for custom applications. In the most recent use case, the team leveraged the imperative side of Spark to redesign a complex pipeline for large-scale language model training, which led to reductions in both resource usage and data landing time.
Large-scale monitoring systems generally cannot handle large-scale analysis in real time because the query performance is too slow. To address this, we developed and recently open-sourced Beringei, a high-performance in-memory time series database. Beringei currently stores up to 10 billion unique time series and serves 18 million queries per minute, powering most of the performance and health monitoring at Facebook, while enabling our engineers and analysts to make decisions quickly with accurate, real-time data.
Online search has traditionally been a text-driven technology, even for photos and videos. Today, we announced a search system, available in the US, that leverages image understanding to surface the most relevant photos quickly and easily. Using cutting-edge deep learning techniques to process billions of photos and understand their semantic meaning, people can find photos from their friends based on the image content instead of relying on tags or surrounding text.
Every day, more than a billion people use Facebook on mobile devices. Securing data in transit between our mobile apps and our servers helps ensure that people have a safe experience on Facebook. Over the past year, we've built and deployed Zero protocol, an experimental 0-RTT protocol over TCP based on QUIC's crypto protocol. We've seen performance improvements such as a 41 percent reduction in connection latency and 2 percent reduction in total request time, and have contributed some of our findings to TLS 1.3.
Facebook has been transitioning all of its data centers to IPv6-only infra, but the vast majority of people on Facebook only have access to IPv4 internet. Read more about the fix that allowed us to evolve our data center technology while still supporting our entire global community.
Journey inside our data centers to learn how Facebook built some of the most innovative and efficient infrastructure in the industry.
A huge thank you to our community of engineers and developers that continue to make Facebook's open source program a success! We look forward to building more together in 2017!
Our recently launched Recommendations feature, which turns text comments into a dynamically updating map attachment, was the result of collaboration among several teams here at Facebook. Machine learning helps recognize when a post is seeking recommendations and extracts the relevant terms, which are then fed into our local search engine to identify the places with the best match. Our Android and iOS teams also worked to process the information directly on the mobile device to update the post with new recommendations in real time.
Starting today, css-layout (our open source cross-platform implementation of Flexbox) will be available as Yoga, a standalone tool that allows engineers to build layouts quickly for multiple platforms. We implemented Yoga in C for better performance, and included bindings for several languages. You can check it out on GitHub now.
When we built the Reactions feature on Facebook, we wanted it to be a high-quality, lightweight experience. This led us to develop Keyframes, a library that exports After Effects animations using the least amount of data necessary to be rendered back on the client. Today, we're open-sourcing Keyframes so others can create even more fun experiences.
