Showing posts with label SmartNICs. Show all posts
Showing posts with label SmartNICs. Show all posts

Wednesday, September 30, 2020

VMware's Project Monterey for SmartNICs - Pensando's perspective

Pensando Systems is working with VMware on Project Monterey to integrate the next generation of SmartNIC technology into fully virtualized enterprise networks.

The project aims to rearchitect VMware Cloud Foundation to enable disaggregation of the server including extending support for bare metal servers, thereby allowing physical resources to be dynamically accessed by applications based on policy or via software API.

In this video, Silvano Gai of Pensando Systems talks about Project Monterey and the benefits for customers, including performance, scale, and features such as firewalling, micro-segmentation, encryption, telemetry, etc.

https://youtu.be/74BBdbAI9tI



Next-Gen Infrastructure Acceleration

The need for infrastructure acceleration has never been so acute. Advances in silicon are making possible highly programmable and very efficient network flow path engines for true data-centric computing.

Our Next-Gen Infrastructure Acceleration series collects and curates thought leadership videos from top players in this space, including Digital Realy, Evoque Data Center Solutions, NVIDIA (Mellanox), Fungible, Vapor IO and others.

We also present our 2020 Next-Gen Infrastructure Acceleration Report, which discusses infrastructure acceleration technologies adopted by CSPs and explores the products available from leading vendors. The report is available as a free download upon registration.


https://nextgeninfra.io/infrastructure-acceleration/


Pensado emerges from stealth, led by "MPLS" team from Cisco

Pensando Systems, a start-up based in San Jose, California, emerged from stealth to unveil its first product -- a software-defined edge services platform that was developed in collaboration with the world’s largest cloud, enterprise, storage, and telecommunications companies. Pensando ("thinking" in Spanish) is led by Cisco’s legendary “MPLS” team — Mario Mazzola, Prem Jain, Luca Cafiero, Soni Jiandani and Randy Pond. Hewlett Packard Enterprise...


Wednesday, September 16, 2020

Xilinx samples its own telco accelerator card

Xilinx has begun sampling its own T1 Telco Accelerator Card for O-RAN distributed units (O-DUs) and virtual baseband units (vBBUs) in 5G networks. 

The card, which uses the same field-proven Xilinx silicon and IP already being deployed in 5G networks, comes in a multi-function PCIe form factor card which performs both O-RAN fronthaul protocols and layer 1 offload. The card provides advanced workload offload capabilities, enabling a dramatic reduction in the number of CPU cores required in a system. The T1 card also enables the O-DU to deliver greater 5G performance and services while reducing overall system power consumption and cost compared to competitive offerings.

The T1 card is a small form factor, single-slot card that can be plugged into standard x86 or non-x86 servers to achieve the real-time protocol processing performance required for 5G virtualized O-DU platforms. In addition, it offloads line-rate and compute-intensive functions including: channel encoding/decoding using hardened LDPC and Turbo codecs, rate matching/de-matching, HARQ buffer management, and more, freeing the processor cores for running other services - the true promise of virtualization. The T1 card simplifies 5G deployments by offering a turnkey solution through ecosystem partners that includes both O-RAN fronthaul and 5G NR layer 1 reference designs, as well as pre-validated software to enable operators, system integrators, and OEMs to get to market quickly.

Xilinx says the offloading of critical channel coding functions from the CPU to the T1 card delivers up to 45x encoding and 23x decoding throughput improvement relative to the same server without acceleration. The T1 card enables the use of fewer CPU cores, driving down system cost and overall power consumption. Additionally, for O-RAN fronthaul termination, it can process multiple sectors of 5G NR 4TRX at 100 MHz OBW with its 50 Gbps of optical ports. The Fronthaul and L1 bandwidths are matched for optimal scalability; the more towers you want, the more cards you add to the server.

“The trend toward network virtualization and O-RAN has given us an opportunity with the Xilinx T1 Telco Accelerator Card to drive the next steps of disaggregation of standard networks, enabling our expansion into every corner of the 5G market,” said Dan Mansur, vice president of marketing, Wired and Wireless Group, Xilinx. “Working closely with our ecosystem partners, Xilinx hardware, IP and software are leading the innovation and realization of 5G O-RAN networks.”


Wednesday, September 9, 2020

UCF’s OpenSNAPI project gains I/O extension

The Unified Communication Framework (UCF), in collaboration with Arm, announced an open source contribution of an OpenSHMEM-based I/O research extension to access persistent memory storage.

The contributed software enables Smart Networking Adapters to provide real-time access to large datasets and deliver higher application performance for latency-sensitive applications such as fraud detection, cybersecurity analysis, web-scale personalization, and Internet of Things (IoT).

UCF is a collaborative effort by industry, laboratories, and academia to create production-grade communication frameworks and open standards for data-centric and high-performance applications.

“We’ve seen the efficiency gains achieved by offloading network processing to smart adapters, but now we’re experiencing the incredible flexibility and performance available for other offload activities, such as persistent memory storage,” said Brent Gorda, senior director of HPC, Infrastructure Line of Business, Arm. “As an active open source contributor, Arm is pleased to provide to the UCF’s OpenSNAPI project this I/O extension to fuel the next wave of distributed computing applications.”

OpenSNAPI is a collaboration between industry, laboratories and academia with the goal to create a standard application programming interface (API) for accessing the compute engines on the network, and specifically on the smart network adapter. OpenSNAPI allows application developers to leverage the network compute cores in parallel to the host compute cores for accelerating application runtime, and to perform operations and processing closer to the data.

“The UCF’s OpenSNAPI project is helping to expand the applicability and portability of emerging use-cases for smart networking and computational storage to enhance supercomputing performance, offload security or virtualization functions, increase storage performance, and more,” said Steve Poole, UCF board and founding member. “Through open source collaboration with Arm and EMC3 at Los Alamos National Laboratory, the UCF’s OpenSNAPI project is successfully showcasing the flexibility, performance and value of a new class of processing power available in the network.”

https://www.ucfconsortium.org/projects/opensnapi/

Sunday, August 30, 2020

Perspective: Implications of the Huawei Ban

by Brian Klaff, Marketing Director, Ethernity Networks

The banning of Huawei equipment from the 5G core networks of operators in the U.S., the U.K., and a number of other countries across the world is helping to shake up the industry and raise questions about 5G’s deployment future.

Leaving aside the politics, trade disputes, Coronavirus issues, and cybersecurity concerns, it’s worth taking a look at how we got to where we are today from a networking standpoint. How did operators become so dependent on Huawei, and what are their alternatives? What does all this mean for 5G and its users?

Huawei is an ASIC (applied specific integrated circuit) manufacturer, offering these ASIC-based appliances throughout the telecom broadband network. ASICs offer excellent performance at low up-front cost, and Huawei was known as a provider of end-to-end ASIC-based systems at especially low prices.  That drew many telecom operators to Huawei as their primary hardware provider.

Because 4G telecom networks are based on a traditional, monolithic infrastructure, the network core does the heavy lifting, delivering all the bandwidth necessary to run today’s end user applications. As such, it made sense for operators to rely heavily on a single primary hardware provider delivering high performance for a reasonable price.

There is certainly a big downside to this approach.  Huawei’s network is proprietary with no interoperability with other companies’ hardware. It’s an all or nothing decision when it comes to using Huawei equipment.

There is also a price to pay in the long run for choosing an ASIC-based system (not just Huawei’s).  ASICs are limited to their initial programming and must be replaced after field deployment every time there is a new protocol, security algorithm, or feature that becomes indispensable. As British operators are finding out, replacing field-deployed hardware is a proposition that is extremely expensive. So the low up-front cost of choosing Huawei can have steep long-term repercussions.


Why Huawei became problematic from a networking perspective
What worked for 4G isn’t necessarily the best in 5G.  Whereas 4G relied so much on performance from the core, 5G seeks higher bandwidth and lower latency by moving much of that performance to the edge of the network.  The 5G specification calls for a more open, disaggregated network, one that performs under the varying circumstances of the network edge with the ability to connect different elements of the network from different vendors.  When one company maintains so much control over an operator’s network and its security, there is reason for concern. Huawei’s vendor-locked monolithic offering scares politicians, and maybe even the telecom operators themselves.

As operators seek to take back control of their networks by diversifying their network hardware providers, the trend is to eschew Huawei in favor of alternatives.  Even in China, the three major operators have committed to more open networks, and they are also weaning themselves off Huawei in certain areas of their 5G deployments, for example by initiating the Open UPF program.

Operators’ options

Perhaps the easiest option for telecom operators is to swap in another ASIC manufacturer, such as ZTE, Samsung, Nokia, or Ericsson, for Huawei.  This will guarantee similar performance and be relatively inexpensive in the short term, but it doesn’t solve the issue of closed, inflexible networks.
One possible solution is virtualization, which has already overhauled data centers and can ensure the flexibility to choose functionalities and features from a wide range of software providers using standard, off-the-shelf commodity hardware.  NFV (network function virtualization) has been promised for many years already, but with 5G it is becoming a necessity in telecom.

By implementing networking and security functions through software instead of rigid hardware, operators gain the flexibility to choose the best-in-class solution for each network component regardless of vendor. They also gain the programmability to easily adapt to new protocols, algorithms, and features.  Software providers such as MetaSwitch, Mavenir, AltioStar, and Affirmed can offer NFV solutions to be run on CPUs on commercial off-the-shelf servers.

Even some of the traditional ASIC manufacturers, such as Toshiba, Nokia, and Ericsson, have recognized this need for agility.  They have started to embrace Open RAN (Radio-Access Network), and they are creating software-based solutions to address that specific segment of the 5G network.

The problem with a software-only approach is that the software runs on CPUs, which were designed to handle compute and control functions, not intensive networking and security functions.  As such, the performance of software-only networking solutions suffers greatly compared to ASICs, and it takes dozens of CPU cores to achieve similar performance.  This becomes exceedingly expensive, and even worse, it requires a lot of physical space and power, valuable commodities at the edge of the network.

A better option, one that combines the best of both worlds, is to opt for disaggregated, open networks using FPGA SmartNICs to handle the networking and security functions.  FPGAs (field-programmable gate arrays) are hardware processors especially efficient in handling many data processing tasks in parallel that are reprogrammable after being field-deployed.

By incorporating an FPGA onto a network adapter, it becomes possible to offload CPU-intensive data processing functions to optimized hardware while maintaining the flexibility of programmable software solutions. This gives operators ASIC-like performance and NFV-like agility in a compact network card that fits into a commercial off-the-shelf server. It reduces the number of required CPU cores to gain significant savings in capital expense, physical space, and power at the network edge.  For example, Ethernity Networks offers an FPGA SmartNIC that includes a complete router-on-NIC that reduces CAPEX costs on 5G User Plane Functionality (UPF) components by up to 80%.

What this means for 5G

A dirty little secret that the telecom industry doesn’t want you to know is that the 5G rollouts most local operators have been touting for the past year or so haven’t really been true 5G.   They have been what is commonly referred to as “4G Evolved,” applying some 5G principles to the 4G infrastructure to produce better-than-4G performance – but not yet reaching the level that 5G promises.

Operators were relying on their Huawei (or other ASIC-based) network equipment without fully committing to the necessary changes to bring about the 5G revolution.  In that regard, the Huawei ban represents a golden opportunity to hasten the implementation of true 5G networks.

5G infrastructure is a green field, with virtually no carry-over from legacy equipment.  As such, operators can now consider how to replace Huawei with an eye to the future, with less concern about the expense involved. 

True 5G is coming. There is no doubt.   Europe and North America lag behind Southeast Asia in mass 5G deployments, but there is still time to make the hard decisions that will determine the makeup of the networks.  The easy and less expensive route – simply turning to another ASIC-based solution provider – will waste the opportunity to disaggregate and bring out the full potential of 5G.

Instead, telecom operators should consider their options thoroughly and choose their vendors based not only on up-front cost, but on the goal of a network that is high-performance, flexible, long-term cost-efficient, and future-proof. To avoid the possibility that they will need to replace field-deployed 5G components in the next few years, they should opt for an open and programmable solution now.

Effects on end user experience

It has been suggested that only an ASIC provider such as Nokia or Ericsson can replace what Huawei offered in terms of local exchange and street cabinet broadband equipment, but that is not true.  FPGA-based solutions are ideal for this as well, offering high networking and security performance with flexibility to accommodate various protocol configurations, including DSL and passive optical network (PON), which are used for last-mile data transfer.

Moreover, the need to reduce power and physical space is a far more critical requirement at the network edge than in a typical data center, as edge sites have very limited physical space and a fixed power envelope.  An FPGA-based edge appliance can address all the performance and security requirements of broadband aggregation in a compact, power-efficient device.

If operators simply replace Huawei with another ASIC provider, the result will be similar network performance.  But while there may be no deterioration of service, the network would remain closed and rigid.  If they opt for software-only solutions to replace Huawei, they get open, multi-vendor networks but may struggle to achieve similar performance cost-effectively or meet timelines for full deployment.  This could lead to a deterioration of service or higher costs for end users.

But if providers choose solutions that rely on FPGA SmartNICs they can achieve both high performance and open, agile networks.  This would maintain quality of service, lower costs for both operators and subscribers, and shorten time-to-market.  That is why in China –  which is significantly ahead of North America and Europe in 5G deployment – all three of the primary mobile operators are insisting on FPGA SmartNIC solutions for their 5G UPF implementations.

Brian Klaff is the Marketing Director at Ethernity Networks.  With over 20 years of experience, Brian has concentrated on product marketing for the networking hardware industry since 2013, with special emphasis on the telecom sector. Prior to Ethernity, he held senior communications positions at Mellanox and Amdocs.

Wednesday, August 19, 2020

Video: Scaling-out Data Centers with the Fungible Data Processing Unit

Fungible was founded in 2015 to revolutionize the performance, economics, reliability, and security of scale-out data centers.

In this video, Pradeep Sindhu, Co-Founder and CEO of Fungible, shares observations about scale-out data centers and the key innovations of Fungible’s Data Processing Unit (Fungible DPU™) which has been positioned as the “third socket” in data centers, complementing the CPU and GPU.

https://youtu.be/spJAOn_y21A



Tuesday, August 18, 2020

Fungible announces its DPU for scale-out data centers

Fungible, a start-up based in San Jose, California, unveiled its Fungible Data Processing Unit (Fungible DPU), a microprocessor optimized for data interchange and data-centric computation in scale-out architectures.

Fungible describes its DPU as the "third socket" in data centers, complementing the CPU and GPU, and delivering significant gains in performance, footprint and cost efficiencies for next-generation, scale-out networking, storage, security, and analytics platforms.  The company cites two core innovations that are tightly interwoven:

  • A programmable data-path engine that executes data-centric computations at extremely high speeds, while providing flexibility comparable to general-purpose CPUs. The engine is programmed in C using industry-standard toolchains and is designed to execute many data-path computations concurrently. 
  • A new network engine that implements the endpoint of a high-performance TrueFabric that provides deterministic low latency, full cross-section bandwidth, congestion and error control, and high security at any scale (from 100s to 100,000s of nodes). The TrueFabric protocol is fully standards-compliant and interoperable with TCP/IP over Ethernet, ensuring that the data center leaf-spine network can be built with standard Ethernet switches and standard electro-optics and fiber infrastructure.

Fungible is launching two versions of its DPU:
  • Fungible F1 DPU – an 800Gbps processor designed specifically for high performance storage, analytics and security platforms.
  • Fungible S1 DPU – a 200Gbps processor optimized for host-side use cases including bare metal virtualization, storage initiator, NFVi/VNF applications and distributed node security.
Fungible also provides a full suite of software that enables the Fungible DPU and the products it powers to be used "out of the box". This includes data-path stacks, host drivers and agents for x86, and a set of centralized cluster services that provides management, control and visibility of a large number of Fungible DPU-enabled products.

"The Fungible DPU is purpose built to address two of the biggest challenges in scale-out data centers – inefficient data interchange between nodes and inefficient execution of data-centric computations," said Pradeep Sindhu, CEO and Co-Founder of Fungible. "Data-centric computations are increasingly prevalent in data centers, with important examples being the computations performed in the network, storage, security and virtualization data-paths. Today, these computations are performed inefficiently by existing processor architectures. These inefficiencies cause overprovisioning and underutilization of resources, resulting in data centers that are significantly more expensive to build and operate. Eliminating these inefficiencies will also accelerate the proliferation of modern applications, such as AI and analytics."

https://www.fungible.com/


Fungible raises $200 million for Data Processing Units (DPUs)

Fungible, a start-up based in Santa Clara, California, closed $200 million in Series C financing for its efforts to create an entirely new category of programmable processor.

The Fungible Data Processing Unit (DPU) aims to deliver an order of magnitude improvement in the execution of data-centric workloads. The company sees its DPU as a fundamental building block for next-generation data centers.

Fungible has not yet announced its first products. In previous blog posts, Fungible has talked about Composable Disaggregated Infrastructure (DCI), where compute and storage resources are stored in separate servers and interconnected by a very high bandwidth, reliable and low-latency IP over Ethernet (IPoE) network fabric.

Fungible was founded by Pradeep Sindhu and Bertrand Serlet. Sindhu previously founded Juniper Networks, held roles as CEO and CTO, and is now chief scientist. Serlet previously founded a storage startup and before that was senior vice president of software engineering at Apple.

The latest funding was led by the SoftBank Vision Fund with participation from Norwest Venture Partners and existing investors, including Battery Ventures, Mayfield Fund, Redline Capital and Walden Riverwood Ventures. Fungible has raised $300 million to date.

Wednesday, April 1, 2020

Napatech accelerates virtualization with FPGA-based SmartNICs

Napatech released new Link-Virtualization software that delivers high-performance virtual switching capabilities for its family of FPGA-based SmartNICs. The solution delivers a massive increase in network performance and simultaneous reduction in server CPU utilization, allowing data center operators to fulfill their vision for SDN and NFV.

The company said its Link-Virtualization software and FPGA-SmartNIC solution delivers the high throughput and virtual switching programmability needed for use cases such as VM demultiplexing, full and partial vSwitch offload for application acceleration, VM-to-VM network and application performance monitoring, network telemetry, and custom workload integration.

Highlights:


  • Network Performance: More than 60X gain compared to basic NICs with line-rate forwarding up to 100Gbps switching and 60 million packets per second to virtual machines, for millions of simultaneous flows, with sub-10 uS latency.
  • Server Scalability: More than 90% reduction in server utilization, removing burdensome network and security processing from valuable and expensive CPU cores, returning them to the applications, services and tenants for which they were intended.
  • Feature Velocity: Reconfigurable and programmable processing to retain hardware performance and the speed of software innovation, to keep pace with the rapid evolution in networking standards and increasing threat landscape in cybersecurity.
  • Data Center Sustainability: The combination of performance and programmability provides up to 5X lifetime cost savings by requiring fewer servers, rack space, power and cooling, while also reducing costly upgrade cycles when basic NICs fall behind in functionality.
  • Rich Functionality: Napatech’s Link-Virtualization is supported on Napatech’s family of FPGA-based SmartNICs that include 10, 25, 40 and 100 gigabit Ethernet. It is fully compliant with industry standards including DPDK, OVS, and PCIe, and features the industry’s most programmable application interface for sophisticated match-action policies, VirtIO with live migration support, QoS, load balancing, overlay encapsulations, and much more.

“As the market shifts towards virtualized cloud, telco 5G, edge and enterprise networks, Napatech’s programmable NICs and software are leading the way,” said Jarrod Siket, Chief Marketing Officer at Napatech. “No vendor has shipped more FPGA-based SmartNICs and ports than Napatech, and our Link-Virtualization software makes us well positioned in this high growth market.”

The software is in trials with global server and appliances OEMs.

https://www.napatech.com/products/link-virtualization-software/