OCP 2019: Facebook rethinks data center fabric

At this week's OCP Summit in San Jose, Facebook released details on how they're rethinking the fabric of their data centers. FBOSS is still used to bind together its data center, but there are significant changes to ensure that a single code image and the same overall systems can support multiple generations of data center topologies and an increasing number of hardware platforms.

Facebook's next-generation "F16" data center fabric design offers 4x the capacity of its predecessor while promising to be more scalable and simpler to operate and evolve. The fabric leverages commercially available 100G CWDM4-OCP, which yields desired 4x capacity increase as 400G link speeds, but with 100G optics.

The refreshed fabric includes Minipack, a new modular block switch Facebook developed with Arista Networks. The Minipack switch easily integrates into various data center topologies while also consuming half the power of its predecessors. Facebook calculates that Minipack will consume 50 percent less power and space than its predecessor. Its modularity enables it to serve multiple roles in the new topologies.

In addition to Minipack, Facebook also jointly developed Arista Networks’ 7368X4 switch.

Both Minipack and the Arista 7368X4 are being contributed to OCP, and both run FBOSS.

Facebook has also developed HGRID as the evolution of Fabric Aggregator to handle the doubling of buildings per region.

https://code.fb.com/data-center-engineering/f16-minipack/

Read More

OCP 2019: Microsoft's Project Zipline offers better data compression

At OCP 2019, Microsoft unveiled Project Zipline, a new compression standard for data sets covering Edge to Cloud app.

Project Zipline promises "compression without compromises where always-on procession achieves high compression ratios with high throughput and low latency. Zipline encompasses algorithms, software, and silicon engines.

Microsoft estimates Zipline data set sizes at 4 ~ 8% of uncompressed sizes. Over time, Microsoft anticipates Project Zipline compression technology will make its way into network data processing, smart SSDs, archival systems, cloud appliances, general purpose microprocessor, IoT, and edge devices.

A number of big name silicon and storage companies are already on board as partners.

https://azure.microsoft.com/en-us/blog/hardware-innovation-for-data-growth-challenges-at-cloud-scale/

Read More

OCP 2019: Arista's 12.8Tbps switch developed with Facebook

At OCP 2019, Arista Networks announced a high-radix 12.8Tbps switching system developed with Facebook with the goal of simplifying 100/400G networking.

The Arista 7360X Series doubles system density while reducing power consumption and cost by doubling the network diameter and reducing the number of required leaf-spine tiers. Full manageability via FBOSS (Facebook Open Switching Software) is supported for controlling power and thermal efficiency along with the control plane.

The new platform is a compact, four rack unit design and all active components are removable. It delivers a 60% reduction in power at under 10 watts per 100G port. Standards-based, the system comes with support for 100G QSFP and 400G OSFP or QSFP-DD optics and cables. Arista EOS delivers the advanced traffic management, automation and telemetry features needed to build and maintain modern cloud networks.

The Arista 7368X4 Series is available as an 8-slot modular system with a choice of 100G and 400G modules based on industry-standard interfaces and support for EOS.

It is currently shipping with 100G interfaces. Price per 100G port is under $600.

“The Arista solution has helped Facebook to gain significant improvements in power and space efficiency, reducing the number of switch chips in the network stack and allowing power to be freed up for compute resources,” said Najam Ahmad, Vice President Network Engineering for Facebook. “Having both an internally developed Minipack and the Arista solution allows Facebook to remain multi-sourced, with an option to run Arista EOS or FBOSS on both, where either system can be deployed in multiple tiers of networks.”

Read More

OCP 2019: Facebook plans own fiber linking data centers in Ohio, VA, NC

As previously disclosed, Facebook built its own 200-mile underground fiber cable between its data centers in New Mexico and Texas. The cable is described as "one of the highest-capacity systems in the United States, with state-of-the-art optical fiber."

Facebook now plans new fiber builds between its data centers in Ohio, Virginia, and North Carolina.

This blog posting by Facebook's Kevin Salvadori discusses the company's fiber deployments.

https://code.fb.com/connectivity/fiber-optic-cable/

Facebook presentation: Optics Inside the Data Center

Monday, July 09, 2018  Facebook  

Mark McKillop, Network Engineer at Facebook, and Katharine Schmidtke, Sourcing Manager of Network Hardware at Facebook, talk about challenges in Facebook's optical networks, both in backbone and in data centers.

The first part of the video covers the optical systems used to connect Facebook's POPs and data centers.

The second part discusses optical scaling challenges inside the data centers, including the potential for onboard optics in future systems.

This 30-minute video presentation was recorded at Facebook's Networking@Scale 2018 event in June in California.

See video:
https://www.facebook.com/atscaleevents/videos/2090069407932819/

Read More

OCP 2019: New Open Domain-Specific Architecture sub-project

The Open Compute Project is launching an Open Domain-Specific Architecture (ODSA) sub-project to define an open interface and architecture that enables the mixing and matching of available silicon die from different suppliers onto a single SoC for data center applications. The goal is to define a process to integrate best-of-breed chiplets onto a SoC.

Netronome played a lead role initiating the new project.

“The open architecture for domain-specific accelerators being proposed by the ODSA Workgroup brings the benefits of disaggregation to the world of SoCs. The OCP Community led by hyperscale operators has been at the forefront driving disaggregation of server and networking systems. Joining forces with OCP, the ODSA Workgroup brings the next chapter of disaggregation for domain-specific accelerator SoCs as it looks toward enabling proof of concepts and deployable products leveraging OCP’s strong ecosystem of hardware and software developers,” said Sujal Das, chief marketing and strategy officer at Netronome.

"Coincident with the decline of Moore's law, the silicon industry is facing longer development times and significantly increased complexity. We are pleased to see the ODSA Workgroup become a part of the Open Compute Project. We hope workgroup members will help to drive development practices and adoption of best-of-breed chiplets and SoCs. Their collaboration has the potential to further democratize chip development, and ultimately reduce design overhead of domain-specific silicon in emerging use cases,” said Aaron Sullivan, Director Hardware Engineering at Facebook."

https://2019ocpglobalsummit.sched.com/event/JxrZ/open-domain-specific-architecture-odsa-sub-project-launch

Wiki page: https://www.opencompute.org/wiki/Server/ODSA

Mailing list: https://ocp-all.groups.io/g/OCP-ODSA

Netronome proposes open "chiplets" for domain specific workloads

Wednesday, October 31, 2018  Netronome, Silicon  

Netronome unveiled its open architecture for domain-specific accelerators .

Netronome is collaborating with six leading silicon companies, Achronix, GLOBALFOUNDRIES, Kandou, NXP, Sarcina and SiFive, to develop this open architecture and related specifications for developing chiplets that promise to reduce silicon development and manufacturing costs.

The idea is fo chiplet-based silicon to be composed using best-of-breed components such as processors, accelerators, and memory and I/O peripherals using optimal process nodes. The open architecture will provide a complete stack of components (known good die, packaging, interconnect network, software integration stack) that lowers the hardware and software costs of developing and deploying domain-specific accelerator solutions. Implementing open specifications contributed by participating companies, any vendor’s silicon die can become a building block that can be utilized in a chiplet-based SoC design.

“Netronome’s domain-specific architecture as used in its Network Flow Processor (NFP) products has been designed from the ground up keeping modularity, and economies of silicon development and manufacturing costs as top of mind,” said Niel Viljoen, founder and CEO at Netronome. “We are extremely excited to collaborate with industry leaders and contribute significant intellectual property and related open specifications derived from the proven NFP products and apply that effectively to the open and composable chiplet-based architecture being developed in the ODSA Workgroup.”

Read More

OCP 2019: CBTS brings Carrier Open Infrastructure based on OpenCORD

CBTS (formerly Cincinnati Bell Technology Solutions) announced its Carrier Open Infrastructure (COI) reference architecture based on frameworks from the Open Networking Foundation (ONF), open source hardware from the Open Compute Project (OCP) and Open Source VNFs.

CBTS said its goal is to help carriers leverage open source virtual networking functions (VNFs) and common, off-the-shelf (COTS) hardware to grow revenue as broadband speeds increase and average revenue per-subscriber (ARPU) declines.

The COI architecture leverages the ONF's Central Office Redefined as a data center (CORD) framework for enabling gigabit access over copper, fiber and wireless.

CBTS 10GB XGS-PON Access Solutions 
OpenOLT
CO-OLT24XG-PON is a powerful next-generation OpenCORD-compatible 1RU PON access platform designed for remote terminal (RT) and/or central office (CO) applications. It features:

• High-performance processor to ensure device stability and OpenFlow control plane performance
• Interoperability with SDN controllers including OpenDaylight, ONOS and Commercial-source
• 24 Port x XFP XGS-PON + 6 x 100GE ports
• G.9807.1 10G PON MAC
• Up to 256 ONTs/2048 service flows per PON port
• Non-Blocking line rate architecture to forward packet flows at wire speeds on all ports
• Deep packet buffers for high-speed packet processing
• HQoS support
• Service rate limiting for both U/S and D/S.
• Flexibility to define a wide range of match-action table processing (OpenFlow1.3 + multi-table pipelines)
• Guaranteed fast failover (link or device) by supporting large number of flow mod/sec

OpenONU
• XGS-PON SFU (Single Family Unit) XG-99K – ITU-T G.9807 compliant 10GB downstream and upstream XGS-PON interface supports triple-play services including voice, video, and high-speed internet access service
o Compliant with standard OMCI definition, manageable at remote side, supports the full range FCAPS functions including supervision, monitoring and maintenance
• XGS PON ONT SFP XG-99S Plug-in – ITU-T G.9807 compliant XGS-PON interface replaces the Ethernet SFP+ module existing in Ethernet gateway, switch, router and backhaul equipment
o OMCI stack provides all XGS PON functionalities and full range FCAPS management features including supervision, monitoring and maintenance

“We developed COI and our new OCP-based optical access solutions in response to the pressing need we see to support carriers straining to conduct national expansion initiatives in the face of ongoing subscriber demand for higher bandwidth and increased competition driving subscription fees down,” said Greg Harrison, SVP of Service Provider at OnX/CBTS. “These developments build on our long history of success in SDN projects for the world’s largest carriers and on our deep commitment to industry open source initiatives including ONF and OCP. We look forward to continuing to build on the momentum we have created with COI, and to innovating even further with the help of our growing community.” 

https://www.cbts.com/carrier-open-infrastructure/

Read More

OCP 2019: Juniper integrates with SONiC

Juniper Networks will offer native integration of its platforms with Software for Open Networking in the Cloud (SONiC), which was developed and contributed to the Open Compute Project (OCP) Foundation by Microsoft.

SONiC is an extensible network switch operations and management platform with a large and growing ecosystem of hardware and software partners.

Juniper said native integration with SONiC underscores its commitment to open programmability, complete disaggregation and expanded solutions to support cloud-first enterprises. Specifically, the integration will offer cloud and service provider customers:

• Open programmability: Allows for the rapid integration, agility and flexibility necessary for enterprise end users looking to swiftly adapt to market changes.
• Disaggregation: Highly modular architecture decouples integrated components and software, thereby offering customers the ultimate freedom of choice and flexibility.
• Automation: Network operations have always been a tedious and repetitive process. Combining the power of open programmability and disaggregation, Juniper streamlines network diagnostics, automates complex workflows and optimizes network infrastructure operation.
• Broad ecosystem: Native SONiC integration will provide the broad networking community and cloud providers with the latest routing, switching and analytics solutions from Juniper.

“At Juniper Networks, we recognize how important open programmability is to our customers, already evidenced in our support of OpenConfig, Open/R and P4. To continue this support, we’re excited to announce the native integration of Juniper’s platforms with SONiC to offer hyperscale data center customers another option in data center architecture,” stated Manoj Leelanivas, Chief Product Officer, Juniper Networks.

“The integration of Juniper’s platforms with SONiC shows the company’s commitment to open networking and is an important step in our mission to revolutionize networking for today and into the future. Customers will be able to take advantage of this simplified and automated switch management platform, enhanced by rich routing and deep telemetry innovations,” stated Yousef Khalidi, CVP, Azure Networking, Microsoft Corp.

Read More

OCP 2019: Big Switch demos SONiC + Open Network Linux

At OCP 2019, Big Switch Networks demonstrated an open-source network operating system (NOS) through an integration with Microsoft-led Software for Open Networking in the Cloud (SONiC) and Big Switch-led Open Network Linux (ONL). The demonstration highlights automation, zero-touch provisioning and visibility leveraging a DevOps-centric Ansible workflow and SDN-centric controller workflows.

The SONiC + ONL NOS is comprised of the following open-source software components, each of which are widely deployed independently:

• ONL, a base platform OS, including ONLP platform APIs
• SONiC, higher-layer NOS stack, including forwarding agent/Switch Abstraction Interface (SAI) management, telemetry and programmable API layers
• Free Range Routing (FRR), integrated through SONiC, for the L3 control plane functionality (BGP, OSPF)

The SONiC + ONL demo stack is available for download from the SONiC + ONL technology page. (below). Examples from the demo include:

• Configuration automation and visibility with Ansible
• Zero-touch installation and visibility via an SDN controller
• Ease of deploying a BGP switching fabric with 10G, 25G and 100G open networking switches from Edgecore Networks, leveraging Broadcom’s StrataXGS Trident II and StrataXGS Tomahawk networking ASICs

https://www.bigswitch.com/solutions/technology/open-network-linux/onl-sonic

Read More

OCP 2019: Apstra shows data center automation with SONiC

At OCP 2019 in San Jose, Apstra demonstrated its Intent-Based Data Center Automation with support for SONiC (Software for Open Networking in the Cloud) as part of a multi-vendor solution.

Apstra Intent-Based Data Center Automation enables vendor-agnostic data center lifecycle management.

“Apstra is a strong advocate of network disaggregation. Our mission is to deliver massively automated turnkey infrastructure operations that are hardware independent, and we continue to add new network operating systems to our hardware compatibility list based on customer demand,” said Mansour Karam, CEO and Founder of Apstra. “We are showing a demonstration of Apstra Intent-Based Data Center Automation with SONiC, an open source container-based cloud networking stack solution, that represents expanded options for companies deploying Intent-Based Networking solutions.”

Read More

OCP 2019: Broadcom intros OCP NIC 3.0 adapters

Broadcom introduced OCP NIC 3.0 adapters supporting the full range of data rates and interfaces from 1GbE to 200GbE. The portfolio includes a wide selection of Ethernet adapter cards with 1-, 2- and 4-port configurations.

The new 100GbE and 200GbE adapters, which are based on Broadcom's NetXtreme E-Series Ethernet controllers, also feature Broadcom’s Thor multi-host controller that has the industry’s best performing 56G PAM-4 SerDes and PCIe 4.0 interface. Sampling is underway.

“OCP NIC 3.0 provides a unified specification and form factor for connecting server and storage systems,” said Ed Redmond, senior vice president and general manager of the Compute and Connectivity Division at Broadcom. “With strong customer demand and virtually unanimous industry backing for this unified solution, our complete portfolio of OCP NIC 3.0 adapters facilitates broad adoption of this new form factor and drives further innovation in high performance computing and storage applications to address an ever-increasing demand for bandwidth.”

Read More

Determined AI emerges from stealth with focus on deep learning software

Determined AI, a start-up based in San Francisco, emerged from stealth to unveil its plans to offer specialized software that addresses the challenges faced by deep learning.

Determined AI has raised $11 million in series A funding led by GV (formerly Google Ventures), Amplify Partners, CRV, Haystack, SV Angel, The House, and Specialized Types.

The company says its software will feature a DL-aware scheduling system that allows for cluster sharing, fault tolerance, and workload elasticity with sub-second latency. Another goal is to liberate DL investments from the risk of cloud or hardware lock-in.

Determined AI's software has been running on production GPUs for more than a year.

https://determined.ai/

Read More