Since joining Dell as CTO for Networking & Solutions in June 2019, I have been energized by the opportunities and the extent of technology development at Dell Technologies, as well as the deep partner engagement in R&D. Heading into 2020, our customers require distributed and automated infrastructure platforms that support a wide range of use cases from data center automation to edge and 5G enterprise verticals. Let’s take a closer, more technical look at what’s behind these trends.
Cloud-native software drives intelligent automation and fabrics in data centers
Advances in infrastructure automation are leading to full automation stacks incorporating OS configuration management, DevOps tools, and platform stack installers and managers. These bundles enable a new operational model based on fully-automated, zero-touch provisioning and deployment using remote tools for networking, compute and storage infrastructure. This has become a critical requirement for large deployments, delivering the ability to rapidly deploy and manage equipment with the least amount of operational cost at scale. This is a key enabler for edge use cases.
Network configuration and fault mitigation is rapidly becoming automated. Telemetry data availability and integration with orchestration applications allows the network to be more than one static domain. Using data analysis and fault detection, automatic network configuration and self-healing can become a great differentiating factor in selecting one solution over another.
The tools for infrastructure lifecycle management, including firmware upgrades, OS updates, capacity management and application support, are becoming an integral part of any infrastructure solution. These trends will accelerate with the help of AI software tools this year and continue to expand to every part of the infrastructure.
Micro-services based NOS design fuels the next wave in Open Networking
Network operating systems (NOS) are evolving into flexible cloud-native microservices designs that address many of the limitations of traditional networking platforms. One of the biggest benefits is the ability to support different hardware platforms and customize the services and protocols for specific deployments. Gone are the days when the only option network operators had was to accept a monolithic, generic OS stack with many features that would never be used. This new architecture is critical for supporting edge platforms with constrained CPU and power with targeted networking missions.
Community-based NOS platforms such as SONiC (Software for Open Networking in the Cloud) have the added benefit of accelerating development through a community. SONiC is gaining momentum as a NOS for both enterprises and service providers due to its disaggregated and modular design. By selecting desired containers and services, SONiC can be deployed in many use cases and fit in platforms of many sizes.
The recent increased industry involvement and community creation has placed SONiC on an accelerated path to support more use cases and features. The increased development activity will continue through 2020 and beyond. SONiC has also grabbed the attention of other projects and organizations such as ONF and TIP/Disaggregated cell site gateways. These projects are looking into ways to integrate with SONiC in their existing and new solutions and driving a new set of open networking use cases.
Merchant silicon extends to cover more complex networking requirements
Programmable packet forwarding pipelines, deep buffers, high radix, high line speeds, and high forwarding capacity merchant silicon switches coupled to a new generation of open network operating systems are enabling effective large scale-out fabric-based architectures for data centers. These capabilities will enhance both data center and edge infrastructure, replacing the need for a chassis design or edge routers with custom ASICs. In 2020, for the first time, we expect to see merchant silicon-based network solutions achieve parity with most of the traditional edge and core networking platforms, providing a scale out design that is better aligned to converged infrastructure and cloud requirements.
Programmable silicon/data plane enabling streaming analytics
Programmable data planes are maturing with P4 compilers (as the community approach) and many other available languages for creating customized data pipelines. There is also a growing number of NOSs that support programmable data plane functionality. These new software tools enable the creation of unique profiles to support specific services and use cases, including edge functionality, network slicing, real time telemetry and packet visibility. These powerful new capabilities provide control and AI-based mitigation, as well as customized observability at large scale in real time. Developers have access to the data pipeline and will be able to create new services that are not possible in traditional networking. This is going to be one of the key new trends in 2020.
Storage fabrics using distributed NVMe-oF over TCP/IP solutions
NVMe has emerged as the most efficient and low-latency technology for storage access. NVME-over-Fabric (NVMe-oF) extends the protocol to work across networks using fabric-based networks (Fibre Channel, RoCE, TCP/IP). TCP/IP and RoCE have a clear cost effectiveness advantage with 100GbE being four times as fast as 32GbE FC at about 1/8th of the cost. Between those two protocols TCP/IP emerges as the solid choice due to similar performance, better interoperability and routing, and utilization of lossless networks only where needed. NVMe-oF/TCP transport provides the connectivity backbone to build efficient, flexible, and massive-scale distributed storage systems. The key to unlocking this potential is service-based automation and discovery controlling the storage access connectivity within the proven SAN operational approach and orchestration frameworks extended across multiple local storage networks through both storage services and fabric services federation.
Distributed edge emerging as a requirement for Industry vertical solutions
Emerging use cases at the far edge for analytics, surveillance, distributed applications and AI are driving the need for new infrastructure designs. Key constraints are the operating environment, physical location, and physical distribution giving rise to the need for a comprehensive remote automated operational model. New workload requirements are also driving the design. For example, Gartner predicts that “by 2022, as a result of digital business projects, 75% of enterprise-generated data will be created and processed outside the traditional, centralized data center or cloud*.” New innovations at the edge include converged compute and networking, programmable data plane processors, converged rack-level design, micro/mini data centers, edge storage and data streaming, distributed APIs and data processing. We are at the start of new phase of development of custom solutions for specific enterprise verticals that will drive new innovations in infrastructure and automation stacks.
Wireless first designs are driving new infrastructure platforms for enterprises and service providers
There is tremendous growth in wireless spectrum and technologies including 5G, 4G, shared spectrum (CBRS), private LTE, and WiFi, coupled with a new desire to transition to wireless as the preferred technology for LAN, campus, eetail, etc. This is driving the need for wireless platform disaggregation into cloud native applications for core, radio access network (RAN) and WiFi that support multiple wireless technologies on shared infrastructure. Disaggregation is starting at the core and moving to the edge levering edge compute with automation in a distributed model, which is bringing all the benefits of cloud economics, automation and developer access to wireless infrastructure and creating massive new efficiencies and new services.
Smart NICs are evolving to address massively distributed edge requirements
The new generation of powerful Smart NICs extend the model of simple NIC offload and acceleration by adding heavy data plane processing capacity, programmable hardware elements, and integrated switching capabilities. These elements allow many data flow and packet processing functions to live on the smart NIC, including networking, NVMe offload, security, advanced telemetry generation, advanced analytics, custom application assistance, and infrastructure automation. Smart NICs will be a key element in several valuable use cases: distributed network mesh, standalone intelligent infrastructure elements (e.g. radio controllers), autonomous infrastructure, distributed software defined storage, and distributed data processing. Smart NICs will serve as micro-converged infrastructure extending the range of edge compute to new locations and services beyond edge compute.
The age of 400G – higher speeds driving new fundamental network switch architecture
Native 400G switches coupled with 400G 0ptical modules are now available and breaking the 100G speed limit for data center interconnects. This is creating challenges with power and thermal, as well as space and layout, and moving the industry to co-packed optics.
In addition, new silicon photonics (ZR400 and others) enable long reach Dense Wavelength Division Multiplexing (DWDM) transport given the availability of merchant optics DSPs. This is going to fundamentally transform networking, data center interconnect and edge aggregation by collapsing the need for a stand-alone DWDM optical networks, therefore bringing great efficiencies, automation and software-defined capabilities to the entire networking stack.
Stay tuned—2020 is set to be a year packed with innovation as we strive to deliver customers the technology that will drive their businesses into the future.
Posts
0 comments:
Post a Comment