SlideShare utilise les cookies pour améliorer les fonctionnalités et les performances, et également pour vous montrer des publicités pertinentes. Si vous continuez à naviguer sur ce site, vous acceptez l’utilisation de cookies. Consultez nos Conditions d’utilisation et notre Politique de confidentialité.
SlideShare utilise les cookies pour améliorer les fonctionnalités et les performances, et également pour vous montrer des publicités pertinentes. Si vous continuez à naviguer sur ce site, vous acceptez l’utilisation de cookies. Consultez notre Politique de confidentialité et nos Conditions d’utilisation pour en savoir plus.
July 2016 – facebook Open Cellular
access platform that can support a wide variety of wireless network
standards, from 2G and LTE to Wi-Fi access points. Anyone can
customize the platform to meet their connectivity needs and set up
the network of their choosing, in both rural and urban areas
Sofar able to send and receive SMS
messages, make voice calls, and use basic
data connectivity using 2G implementation …
working with TIP on next steps
Nov 2016 – Facebook announce Voyager
Based on facebook Wedge100 ToR switch
working with a few partners on Open Packet DWDM
— a packet transponder and open line transport
system with open optical specifications that enable
any interested party to contribute systems,
components, or software.
Voyager, a networking solution for Open Packet
DWDM networks —announced as the industry’s first
“white box” transponder and routing solution.
The open line system will include Yang software data
models of each component in the system, and an
open northbound software interface (NETCONF,
Thrift, etc.) to the control plane software. This allows
multiple applications to run on top of the open
software layer, enabling software innovations in
DWDM system control algorithms and network
Facebook Voyager (2)
The Voyager transponder includes the same switch ASIC as Wedge 100 for aggregating the 100 GbE client
signals (Broadcom Tomahawk).
In addition, Voyager implemented the DSP ASIC and optics module (AC400) from Acacia Communications
for the DWDM line side with their open development environment.
Facebook worked with Lumentum to develop a terminal amplifier specification so that multiple applications
can run on top of the open software layer to enable software innovations in DWDM system control
algorithms and network management systems.
Voyager's optical capabilities cover applications from metro to ultra-long-haul reaches.
We successfully tested the packet-optical transponders in field trials with Equinix in the U.S. and MTN in
South Africa, and we plan to open-source the Voyager software, similar to the FBOSS software for Wedge
Nov 10, 2016 – mmwave data rate record
Millimeter Wave Point-to-Point:
record data rate of nearly 20 Gbps over 13 km with MMW
technology. Using a set of custom-built components, the team
achieved this milestone with only 105 watts of total direct
current (DC) power consumption at the transmitter and
receiver. The transmission used a bandwidth of 2 GHz,
resulting in an overall spectral efficiency of 9.8 bits per
second per Hertz.
Test performed in southern California between a mountain in
Malibu and a building 13.2km down
enough data to stream almost 1,000 ultra-high-definition videos at the same time. But still many technical
challenges to solve …
The team is currently flight testing its first generation air-to-ground bidirectional link capable of 20 Gbps in each
direction. The aerial payload is mounted on a Cessna aircraft and is being flown at altitudes up to 20,000 ft.
The next generation air-to-ground communication system capable of supporting 40 Gbps each on uplink and
downlink between an aircraft and a ground station will be flight-tested in early 2017
2002 – WindMill
Windmill, open sourced in 2012:
The Server V2 design, also known as “Windmill,” was a
power-optimized, bare-bones motherboard for Intel Xeon
processors designed to provide lowest capital and
operating costs. It did away with many features that
vendors usually include in servers but that aren’t
necessary for Facebook’s needs.
March’15 – Facebook Open Compute “Yosemite”
first open source modular chassis for high-powered microservers
Contributed to OCP (Open Compute Project)
applies to two-socket (2S) computing platforms, which have become
scale-up systems. 2S has been the mainstream server architecture
for a long time for good reason. With multiple high-performance
processors, it's strong and versatile, but it's also bulky and power-
hungry. In other words, it's not optimized for scale-out uses.
•A server-class SoC with multiple memory channels, which provides high-performance computing in 65W TDP for
SoC and 90W for the whole server card.
•A standard SoC card interface to provide a CPU-agnostic system interface.
•A platform-agnostic system management solution to manage the system and these 4 SoC server cards, regardless
•A multi-host network interconnect card following OCP Mezzanine Card 2.0 specification, which connects up to 4
SoC server cards through a single Ethernet port.
•A cost-effective, flexible, and easy-to-service system structure.
This system will be fully compatible with Open Rack, which can accommodate up to 192 SoC server cards in a
single rack. Already supported by Mellanox
FBOSS – FaceBook Open Switching System
project on GitHub: https://github.com/facebook/fboss
FBOSS consists of a number of Linux user-space
applications, libraries, and utilities.
The initial open source release of FBOSS consists
primarily of the agent daemon, but Facebook is working
on releasing additional pieces and functionality as well.
One of the central pieces of FBOSS is the agent daemon,
which runs on each switch, and controls the hardware
forwarding ASIC. The agent daemon sends forwarding
information to the hardware, and implements some control
plane protocols such as ARP and NDP.
The initial FBOSS agent release is targeted for the Broadcom StrataXGS series of Ethernet switch ASICs
(specifically the Trident and Trident II chips) using Broadcom opensource OpenNSL APIs.
•Programming various tables within the Broadcom ASIC, such as L2, L3, and VLAN
•Handling low-level control packets for host and neighbor learning (ARP, IPv6 NDP,
DHCPv4/v6 relay, LLDP).
•Packet parsing/construction of ICMP/UDP packets.
Baseboard Management Controller (BMC) software stack
a BMC is a specialized controller embedded in servers. It often comes in the form of a system-on-chip
(SoC), with its own CPU, memory, and storage and lots of IO. A BMC connects to sensors to read
environmental conditions and to fans to control temperature. It also provides other system management
functions, including remote power control, serial over LAN, and monitoring and error logging of the server
host CPU and memory.
BMC hardware is a computer system.
Compared with modern computer
systems, the hardware resources in a
BMC are very limited. A slow CPU, less
than 32 MB flash as storage, and less
than 256 MB RAM are common in BMC
hardware. Because of this, OpenBMC is
designed as a complete Linux
distribution, with flexibility to be
customized to support different BMC
SoC or boards.
Ex: “OpenBMC” packages running on the BMC inside “Wedge.”
Facebook and Open Source Security - osquery
osquery allows you to easily ask questions about your Linux, Windows, and OS X infrastructure.
Whether your goal is intrusion detection, infrastructure reliability, or compliance, osquery gives you
the ability to empower and inform a broad set of organizations within your company.
osquery gives you the ability to query and log things like running processes, logged in users,
password changes, USB devices, firewall exceptions, listening ports, and more.
You can perform ad-hoc queries or schedule them, optionally enable FIM and process accounting
CentOS, Ubuntu LTS, Windows, and OS X are supported with no dependencies. osquery powers
some of the most demanding companies, including Facebook.
Facebook ToR Switch (Top of the Rack Switch) Wedge100
Facebook Wedge 100 specification has been accepted into OCP
(Open Compute Project)
May’16 - Terragraph
Facebook Terragraph is a 60 GHz, multi-node wireless
system focused on bringing high-speed internet connectivity
to dense urban areas. Utilizing commercial off-the-shelf
components and leveraging the cloud for intensive data
processing, the Terragraph system is optimized for high-
volume, low-cost production.
Radios are based on the WiGig standard and are designed
for consumer electronics, which allows to create nodes that
are inexpensive relative to traditional telecom infrastructure.
Given the limited range of the 60 GHz signal, these nodes
are placed across a city at 200-250 meter intervals.
Terragraph implements a phase array antenna to retain the
highly directional signal required for 60 GHz, but makes it
steerable to communicate over a wide area.
Terragraph implements IPv6-only nodes, an SDN-like cloud compute controller, and a new modular routing
protocol for fast route convergence and failure detection. We also re-architected the MAC layer to solve the
shortcomings of TCP/IP over a wireless link. By implementing a high performance TDMA-TDD MAC, we saw
up to 6x improvement in network efficiency and at the same time made TCP/IP predictable compared to the
existing Wi-Fi/WiGig standard.
May’16 - Arries
Project ARIES is a proof-of-concept effort to build a test
platform for incredibly efficient usage of spectrum and
- Spectral efficiency: total number of bits transmitted per
unit of radio spectrum bps/Hz, allowing for higher
throughput in even the smallest bandwidths
- Energy efficient : total number of bits transmitted per unit
Joule energy spent b/J
A base station with 96 antennas, it can support 24
streams simultaneously over the same radio spectrum. It
is able to demonstrate 71 bps/Hz of spectral efficiency,
and when complete ARIES will demonstrate an
unprecedented 100+ bps/Hz of spectral efficiency.
So far, Aries demonstrated 1.05 Gbps bidirectional (2.1 Gbps total throughput per distribution node) in P2P
mode, up to 250 meters away. This means up to 8.4 Gbps of total traffic per installation point assuming 4 sectors.
Facebook think this number can be as high as 12.8 Gbps in the future. In P2MP mode, the system is able to
autodiscover the location of the client nodes, and has been able to demonstrate electronically beam-forming the
signals between 2 client nodes in 8 microseconds, or about 125,000 times per second.
May’16 - Facebook deploying Fiber
May 2016 – Facebook, Microsoft & Telefonica join
forces to deploy Fiber across Atlantic
The new project, called Marea, will span the more than
4,000 miles between Virginia and Spain with eight pairs
of fiber optic strands, which would make it the highest
capacity link across the Atlantic - roviding up to 160
terabits per second of bandwidth—about 16 million
times the bandwidth of your home Internet connection
Oct 2016 – Facebook and Google team up to deploy a
8000 miles Fiber between Los Angeles and China
The fiber-optic cable will have a bandwidth of 120
terabits per second, which Google says makes it the
highest-capacity route between the US and Asia
Facebook Machine learning –AI Project
Facebook formed the Applied Machine Learning team
last September 2015. The group runs a company-wide
internal platform for machine learning called FBLearner
Flow. The platform is the artificial-intelligence equivalent
of the Open Compute project with one key difference: It’s
not something that will be offered to the world through
open source hardware. Without the data that Facebook
has on tap, the company says, its platform is essentially
FBLearner Flow combines several machine-learning
models to process several billion data points, drawn from
the activity of the site’s 1.5 billion users, and forms
predictions about thousands of things: which user is in a
photograph, which message is likely to be spam. The
algorithms created from FBLearner Flow’s models &
500,000 workflows help define what content appears in
your News Feed and what advertisements you see.
Joaquin Quiñonero Candela is director of Applied
Machine Learning at Facebook.
Facebook research in AI , Artificial Intelligence
Facebook runs its own AI research lab as well as a Brain-like team known as the
Applied Machine Learning Group. Its mission is to push AI across the entire family
of Facebook products, and according chief technology officer Mike Schroepfer, it’s
already working: one in five Facebook engineers now make use of machine
learning. Schroepfer calls the tools built by Facebook’s Applied ML group “a big
flywheel that has changed everything” inside the company. “When they build a new
model or build a new technique, it immediately gets used by thousands of people
working on products that serve billions of people,” he says.
“We’re trying to build more than 1.5 billion AI agents—one for every person who
uses Facebook or any of its products,” says Joaquin Candela, the head of the newly
created Applied Machine Learning group.
As of March 2016, about 750 Facebook engineers and 40 different product teams
were using the FBLearner Flow platform. By the end of June, the company hopes
that 1,000 engineers will use it. Facebook ultimately aims to build machine learning
tools that are easy enough to use for non-engineers, though that’s a far-off goal.
Facebook AI – FBLearner Flow
– Core concepts:
– Workflows: A workflow is a single pipeline defined within FBLearner Flow and is the entry point for all machine
learning tasks. Each workflow performs a specific job, such as training and evaluation of a specific model.
Workflows are defined in terms of operators and can be parallelized.
– Operators: Operators are the building blocks of workflows. Conceptually, you can think of an operator like a
function within a program. In FBLearner Flow, operators are the smallest unit of execution and run on a single
– Channels: Channels represent inputs and outputs, which flow between operators within a workflow. All channels
are typed using a custom type system that we have defined.
The platform consists of three core components:
• an authorship and execution environment for custom distributed workflows,
• an experimentation management UI for launching experiments and viewing results, and
• numerous predefined pipelines for training the most commonly used machine learning algorithms at Facebook
All workflows and operators in FBLearner Flow are defined as functions in Python and apply special decorators to
integrate with the platform
June’16 - Facebook Internet Drone 1st flight – Aquila Project
goal is to have a fleet of Aquilas flying together at
60,000 feet, communicating with each other with
lasers and staying aloft for months at a time
=> June 2016 – the drone stayed up 96mn
Target: weight < 1000 pounds
power < 5000 W (= 3 hair dryers)
Speed – 80mph
Facebook Population Distribution Study
Applying techniques from computer vision on DigitalGlobe
high-resolution satellite imagery
Created a population data set with 5-meter resolution for 20
Analyzed 23 countries, which amounts to 21.6 million
square kilometers and 350 TB of imagery. For one pass of
our analysis we processed 14.6 billion images with our
convolutional neural nets, typically running on thousands of
Facebook Open Population Data Set project is in
collaboration with Columbia University and the World Bank.
It is starting to share data, globally and per countries.
44% of the population lives in urban areas, while 99% of the population lives within 63 km of the nearest city.
Facebook Open/R – open routing
Open/R generalizes the concept of a replicated state database found in well-known link-state routing protocols
such as OSPF and ISIS. It uses this as an underlying message system upon which we can build multiple
applications. Distributed routing is just one of the applications that leverages this message bus.
Uses mature open source ZeroMQ library for all message exchange, whether it's intra-process or inter-process.
Used as baseline routing technology for facebook Terragraph network
Will be opensourced at some point
Facebook DHCP Load Balancer
Dhcplb is written in Go and can be fetched from GitHub
How it’s used in facebook:
TORs (top-of-rack switches) at Facebook run DHCP
relayers, which are responsible for relaying broadcast
DHCP traffic (DISCOVERY and SOLICIT messages)
originating within their racks to anycast VIPs, one for
DHCPv4 and one for DHCPv6.
Facebook uses ISC KEA opensource DHCP servers
Kea is an open source DHCPv4/DHCPv6 server being
developed by the Internet Systems Consortium (ISC).
Nov 8, 2016 – Backpack Switch, 100Gb Switch
Backpack has a fully disaggregated architecture that uses simple building blocks called switch elements,
and it has a clear separation of the data, control, and management planes. It has an orthogonal direct
chassis architecture, which enables better signal integrity and opens up more air channel space for a better
thermal performance. Finally, it has a sophisticated thermal design to support low-cost 55C optics.
– Created in 2004
– Revenues: $17.9B, net income: $3.7B
– Employees: 15,724
– Users of Facebook social network: $1.79B (Sept 2016)
– Products: facebook, messenger, whatsapp, oculus
– Revenue model: advertising via 3 million advertisers
– Facebook is both a consumer of and contributor to free and open source software. Facebook's
contributions include: HipHop for PHP, Fair scheduler in Apache Hadoop, Apache Hive, Apache
Cassandra, Open Compute Project, and TIP (Telecom Infra Project)
– Facebook Data Centers include: Prineville/Oregon, Forest City, North Carolina, Lulea, Sweden,
and Altoona, Iowa. The company is also building data centers in Fort Worth, Texas - Clonee, Ireland - and
Los Lunas, New Mexico. And is aiming Singapore and Danemark
– Facebook signed an agreement with Greenpeace on environment