Networking/Summit-2017-03

From OpenCompute
Jump to navigation Jump to search

Agenda / Presentations[edit]

We had two full tracks of Networking presentations this year at Summit! Videos of presentations are available here: http://www.opencompute.org/wiki/Summit/US17#Networking_1.

0900 - Welcome to OCP Networking[edit]

Omar and Scott kick off the day with an overview of the OCP Networking group's accomplishments in the past year, the organization of the OCP Networking project and OCP itself, the schedule of the networking workshop talks, and discussion of the Networking project's direction for the coming year.

  • Omar Baldonado, Facebook
  • Scott Emery, Cumulus Networks

[video]

Track 1 (Times in PST)[edit]

0930 - Facebook's Open Networking Hardware.[edit]

This workshop will provide a technical overview of the latest networking devices coming from Facebook's hardware engineering teams: Backpack, their 128x100G modular chassis, and Wedge 100S, a new version of the original 32x100G Wedge 100 switch. The workshop will also include engineers from Cavium and Barefoot Networks, who will talk about the variations of Wedge 100 that they have developed to integrate their programmable ASICs into the platform, and the new use cases that are unlocked by their technologies.

  • Alik Fishman, Cavium
  • Akin Koyuncuogu, Barefoot Networks
  • Xu Wang, Facebook
  • Zhiping Yao, Facebook

[video], [slides]

1030 - Tales From the Field: L1 Issues[edit]

With the rise of Open Networking, where a user is able to switch out any piece of the stack without vendor lock in (switch, pluggable, NOS, etc…), we have only seen more issues arise when compared with vendor locked solutions. Most of these issues happen at L1 between the switch, the NOS, and the pluggable (cable or optic). Even companies that promote open networking solutions and provide internal testing fall victim to L1 issues. This talk is centered around common L1 issues observed from the field and how some vendors have banded together to eliminate them.

  • Carlos Cardenas, Cumulus Networks

[video], [slides]

1100 - Universal Connectivity and Interoperability in the Open Ecosystem[edit]

Interoperability is essential for customer confidence and wide adoption. This talk will address networking interop complexities inherent in an Open Eco system, including some problems that are unique to an Open Eco system, and that must be addressed for successful adoption. We'll examine specific types of Interop Problems found during plugfests and lab testing at UNH-IOL. Finally we'll examine what the community is doing to combat this and ensure interoperability.

  • David Woolf, University of New Hampshire InterOperability Laboratory (UNH-IOL)

[video], [slides]

1130 - ONIE with Secure Boot[edit]

Network security and validating the software running on data center systems has never been more important. The central components for launching trusted operating system environments are the family of technologies described by the Universal Extensible Firmware Interface (UEFI) Secure Boot and the Trusted Computer Group (TCG) Trusted Platform Module (TPM) specifications. This talk discusses the plans of the Open Network Install Environment (ONIE) boot loader project to address the challenges of deploying secure technologies on network switches. The goal is to combine the easy to use image discovery features of ONIE with the industry standard security components of UEFI and TCG. The benefits are:

  • Maintains the established image discovery methods of ONIE
  • Build on industry standard specifications and technologies
  • Support for x86_64 and AARCH64 UEFI firmwares
  • Enables all UEFI capabilities
  • Curt Brune, Cumulus Networks

[video], [slides]

1300 - Linux Networking Greatness, episode 2[edit]

This talk is the second installment of the “Linux is awesome” talk. Linux is undoubtedly the platform of choice for open networking today. Linux kernel and ecosystem provide a flexible and faster development platform for networking applications. In this talk we will look at the latest developments in the Linux kernel and ecosystem for networking in the past year: VRFs, light weight tunnels, enhancements to the traffic classifier, EBPF, XDP (express data path) and more. Mostly centered around running a Linux distribution natively on your open networking gear, in this talk we will cover:

  • New Linux kernel and ecosystem advancements in the past year
  • Example Open Linux networking applications
  • Monitoring your networking apps using systemd
  • Linux Networking configuration
  • Example deployments
  • And more ...
  • Roopa Prabhu, Cumulus Networks

[video], [slides]

1330 - Experience in Operating a Mixed OCP/Non-OCP Data Center Network[edit]

Facebook has been aggressively rolling out its Wedge, Wedge100, and now Backpack devices throughout its data centers. Yet, our data centers are (and will continue to be for quite a while) a mix of these OCP-accepted, disaggregated networking devices running FBOSS and traditional, non-OCP networking devices from traditional networking vendors. The Facebook Net Systems group writes the software on our switches (FBOSS) but also writes the software that provisions, automates, monitors, and controls these mixed data centers. This talk will cover learnings from how we had to adapt our existing systems to handle OCP networking devices such as Wedge/Wedge 100 alongside the traditional networking switches. It will also highlight patterns/conventions that we'd like to see all networking devices move toward to make operations of the network smoother.

  • Starsky Wong, Facebook

[video], [slides]

1400 - Open Networking Advancing, Expanding, and Deploying[edit]

This presentation will cover three topics:

  1. New contributions to Open Compute advancing Open Networking which includes a 64 port 100G switch and two Wi-Fi Access Points based on Qualcomm Wave2 silicon
  2. Demonstrations and presentation on the expansion of CBW Ecosystem partners including ONIE download demonstrations using BRCM and QCA OCP Wi-Fi Access points with NOS solutions from Benu Networks, Mojo Networks, and Canonical. All of this will be powered by AS4610-54P switch running SnapRoute FlexSwitch.
  3. Real word use cases and deployments – 1. Presentation from Major retailor utilizing Open Networking equipment and SW, and 2. Presentation from Major carrier deploying Open Networking equipment and SW
  • Jeff Catlin, Edgecore Networks

[slides]

1430 - Alpha 10G, 25G, and 100G switches for the DC and Telco Markets[edit]

Alpha is sharing the specifications of its switches that expand its OCP product line up to 100G and into the telco market.

  • Robert Chu, Alpha

[video], [slides]

Track 2[edit]

0930 - SAI: Releasing the Potential of Switch ASIC[edit]

Switch Abstraction Interface (SAI) provides simple, consistent APIs for the upper network stack to consume the underlying complex, heterogeneous hardware easily and faster. Multiple popular NOS/network control stack are running on top of SAI, as a fundamental element of disaggregation. We are excited to announce SAI 1.0 release. Architects from major silicon companies will be on the stage to talk about advanced capabilities added/will be added to SAI and use cases lit up through these APIs.

  • Xin Liu, Microsoft

[video], [slides]

1000 - Networking SAI Behavioral Model Implementation in P4[edit]

Providing standard interfaces for managing Ethernet switch device (SAI) is a great step towards HW agnostic NOS. Still, different underlying HW pipelines may result in HW specific product behavior. OCP SAI community has been defining a generic logical pipeline that can represent a common denominator of different Silicon implementation, named SAI behavioral model. Mellanox is implementing this model in P4 language to create a SW switch that is controlled by SAI interface. We would like to present the work, share the repository with the community, discuss obstacles we found along the way, and ask for other member's contributions.

  • Aviad Raveh, Mellanox

[video], [slides]

1030 - Taking SAI Beyond the Data Center Fabric[edit]

In this talk, we’ll examine the applicability of both SAI and SONiC to use cases which go beyond a typical DC fabric and argue that these use cases can drive the broader adoption of these technologies. We’ll look at the functional requirements these use cases place upon both SAI and SONiC; as well as the proof points in terms of scale, robustness and high-availability that both OCP-conformant hardware and software will need to be able to demonstrate to provide a compelling deployable solution for these cases.

  • Alan Elder, Metaswitch

[video], slides

1100 - SONIC: Enabling Fast Evolution in the Network[edit]

Running one of the largest clouds in the world, Microsoft has gained a lot of insight into building and managing a global, high performance, highly available and secure network. We built the these intelligence into Software for Open Networking in the Cloud (SONiC), which thrive for network switch operations and management. Dave Maltz, Distinguished Engineer from Microsoft, will deep dive into the design philosophy underneath SONiC, e.g. containerization, orchestration and how they enable the fast evolution of hyperscale networking.

  • Dave Maltz, Microsoft

[video], [slides]

1130 - Testing to Enable 100G Fabrics at Scale (TOR and Optics)[edit]

This talk covers development testing methods used at Facebook to qualify Wedge100 TOR switch and 100G CWDM4 optics targeted at configuring fabric networks and Network interconnect cards, DAC cable solutions and monitoring used to enable 100G OCP network rack solutions at Facebook. The talk also covers test data and notable anomalies discovered during testing.

  • Abhijit Chakravarty, Facebook
  • Srinivasan Giridharan, Facebook
  • Ashwin Poojary, Facebook

[video], [slides]

1330 - OpenSwitch (OPX) Introduction[edit]

Open Switch is a project dedicated to making a solid Open Source option for a complete Network platform supporting the community and providing networking innovations. Open Switch has evolved recently to with new players Dell and Snaproute stepping in while HP has decided to exit the OpenSwitch project for now. With the changes there is a re-focus on making a completely open source NOS providing both a solid (Linux Based Package approach) allowing the installation of Open Switch on a Debian Jessie based platform. To the OPX project, Dell contributes the Open Source Base directly from their OS 10 Open Edition product while Snaproute has taken the role of providing the control plane software running on top. In this talk we will go through the architecture of the both of these components and describe to the community how and where they can help out.

  • Cliff Wichmann, Dell

[video], [slides]

1400 - Controlling the Network with APIs[edit]

SnapRoute will demo the API-driven model that is used by the OCP project FlexSwitch. The showcase will be how using APIs as the control point of the network can allow for high-levels of automation and programability. Also, the integration of FlexSwitch APIs with other OCP projects such as Open Network Linux and ONIE will be demonstrated.

  • Glenn Sullivan, Snaproute

[video], slides

1430 - 100G Optics Deployment in Facebook Data Centers[edit]

This engineering workshop will describe Facebook’s deployment of 100G optics in switching networks with examples of recent data center use-cases. The workshop will provide details on the optical transceiver specification and the type of fiber installed. Examples will be given of production testing used to ensure these hyper-scale networks deploy successfully, and the focus on quality that ensures they keep running. Details of the ecosystem of optical transceivers and switching equipment that supports 100G deployments will be shared.

  • Abhijit Chakravarty, Facebook
  • Cathie Deal, Facebook
  • Reza Niazmand, Facebook
  • Katharine Schmidtke, Facebook
  • Vincent Zeng, Facebook

[video], [slides]