Are You Ready to Deliver Cloud-Native Network Infrastructures?

Cloud-native principles and technology have already proven to be quite effective in building and operating some of the world's largest networks by leading hyperscalers. Now, as service providers and enterprises race to accelerate their 5G deployments and digital transformations, they are also embracing the cloud-native principles in their network designs. Networking vendors are delivering hyper-converged infrastructure, cloud-native network functions (CNF), and SDN-enabled network orchestration and automation solutions to realize the cloud-era network.

Progress is an evolution that comes with new challenges

With traditional networking, “change” was difficult and time-consuming. About a decade ago, the transition to virtualization started taking place, and networking vendors “lifted and shifted” large monolithic networking functions into virtual machines (VMs). However, we used the same concepts of the past and just made them virtual (vSwitches, vRouters, vFirewalls) and the same mechanisms and protocols, such as virtual extensible LAN (VxLAN), subnets, routing (BGP) to replicate everything from the physical networks into the virtual environment.

This enabled disaggregation from proprietary custom hardware and allowed organizations to reap the cost benefits of commercial off the shelf (COTS) hardware and service chaining to build and deploy customer-facing services from a set of interconnected, best-of-bread multi-vendor VNFs (virtual network functions). However, several challenges slowed the vision for the cloud-era network, including:

• Performance impact due to VM software overhead
• Availability impact due to long VM boot times
• Lack of agility in deployment and configuration of NFV infrastructure (OpenStack)
• Lack of API-driven elastic scalability

With applications and microservices popping up across all domains (cloud regions, on-premises, across the globe), the network connectivity problem for even a single application became much worse under this virtualized network system.

How do we make better virtual networks?

To solve the challenges mentioned, virtualized networking is now going through another evolution to “cloud-native” architectures. Software orchestrated and managed, cloud-native networks break network functions into smaller packages, enabling network operators to deploy and upgrade networks more quickly and deliver on-demand elasticity and resiliency. Typically, this is achieved with the use of Kubernetes to orchestrate a cluster of cloud-native network functions. The cloud-native transformation of networks will be an evolution and here is an excellent pictorial representation of the journey from Dan Meyer of SDxCenteral.

Figure: Cloud-native transformation journey and deployment options - source

Networking vendors are delivering the building blocks to enable these cloud-native networks. If you are a vendor delivering hyperconverged infrastructure or compute platforms for enterprise or telco cloud requirements, you need to be able to help your customers dimension and configure their NFV infrastructure (NFVI) adequately for their workloads. If you are vendor delivery VNFs and CNFs, you need to validate the VNFs and CNFs running on top of various hyperconverged infrastructure (private or public, VMs, or containers) and characterize the impact of compute, network, storage NFVI resources on the performance and user experience of your network functions.

Here are three key tenants to need to keep in mind to void pitfalls; this by no means an exhaustive list but a few key observations from our interactions with customers.

1. Multi-tenancy

With cloud-native networks, network functions share the underlying infrastructure with the necessary abstraction to provide each service or network slice with a private network experience. The network operator is responsible for maintaining and scaling the underlying infrastructure. However, to ensure high performance, vendors developing VNFs and CFNs must perform performance benchmarks in the presence of “noisy neighbors”. We will soon publish a recent case study that showcases the impact of noisy neighbors and how the vendor was able to convince their service provider customer that the issue is not in their VNF but in the NFVI the customer managed. The bottom line is that in the absence of such test benchmarks for the NFVI, vendors spend weeks to troubleshoot and isolate the issue, and in the meantime, must defend their VNF and CNF performance.

2. Elastic scalability

Cloud and container platforms enable the infrastructure to dynamically scale down and scale up the workloads as per the load and demand. The total lifetime of the workload is reduced to hours from the previous span of months or years. This presents a new set of challenges in rightsizing cloud infrastructures and helping your customers optimize cost while minimizing user disruptions to quality of experience (QoE) and ensuring security and application policies are extending in this elastic, dynamic environment. To validate, you need test agents in private/public clouds and VM/container topologies that actively participate in an auto-scale environment while emulating legitimate applications and malicious threat traffic. With such test agents, you can now benchmark the application performance of cloud instances that involve switching, natting, or going through internet gateways to another cloud location (multi-cloud topologies) or on-premises (hybrid topologies). By leveraging test agents that can elastically scale up and down per the policies of the auto-scale group that it is part of you can accurately measure the performance, user experience, and security impact of elastic scale (up or down) during transient periods.

3. Continuous validation

A major benefit of cloud-native networks is agility and flexibility, which means you can make changes to your network on demand and flexibility gives you a lot of choices on compute, NICs, virtual infrastructure managers (OpenStack, VMware), container orchestrators (OpenSwift, Kubernetes), cloud instances, etc. Even a simple upgrade to a newer CPU or new version on Kubernetes CNIs or Linux distribution can have unexpected consequences on performance. Therefore, vendors delivering NFVI or VNF/CNFs need to continuously validate and optimize their infrastructure components to understand the impact of change/updates by comparing baseline performance (before update) with new performance (after update) to ensure a robust continuous integration / continuous development (CI/CD) function.

Keysight’s Cloud-Native Network Validation

Keysight’s Ixia Cloud Peak benchmarks the performance of virtualized network infrastructures by deploying real virtual machine or container workloads on top of hyperconverged infrastructure. Keysight’s CyPerf re-creates every aspect of a realistic application workload across on-premises cloud and container environments to tune the balance between user experience and security on your cloud-native networking solutions.

Supported Use Cases

Related Posts