top of page

Blog Article

Anatomy of a Network Operating System



The IP router is a critical component in the scalability of the internet. A new Network Operating System architecture is needed, one that mimics the reliability and security mechanisms developed by cloud computing platforms. Disaggregated Networking has adopted cloud technologies, therefore it is the ideal solution to scale the internet even more.


According to ourworldindata.org, the number of internet users increased from 413 million in 2000 to over 3.4 billion in 2016 and reached 4.9 billion in 2021.


Every day for the past 5 years, an average of 820,000 people went online for the first time.


Yet, the internet is thriving. We watch 4K videos, make phone calls, visit websites and organize all our business on the internet. For example, videoconferencing has seen a 535% increase in 2020. We have come to heavily rely on the internet for all aspects of our lives.


There is however a fundamental question; how can we make sure that the internet will continue to scale to bring online the next billion users and enable always more sophisticated and bandwidth-hungry applications?


The answer lies in the ability of IP routers to handle the traffic in all parts of the internet network.


The IP router is the fundamental building block of the internet


An IP router has 2 components: the hardware, which forwards the traffic and the Network Operating System (NOS), which mainly computes and applies the routing decisions.


For most people, the performance of a router is only dictated by the capacity of its underlying hardware. However, the NOS has an important role; its architecture will particularly influence the overall performance of the router, which in turn will condition its ability to forward Terabits per second of traffic with little delay.


To make an analogy, you could think about a computer, on which the operating system uses all the hardware resources only for a handful of applications. This may be the result of a monolithic OS architecture, where all functions are centralized, thereby creating a bottleneck in times of peak traffic.


By contrast, cloud computing has come to terms with the issues of scalability, data integrity and availability.


Many of the techniques used by cloud platforms to provide a reliable and scalable service can be transposed to internet routers.


In order to cope with internet traffic growth, a NOS needs to behave like a cloud computing platform. In practice, it means that:


  • It can easily scale capacity. It is not only about forwarding packets from one interface to the other, but also to apply policies, tag packets, filter traffic and much more

  • It has a fault protection mechanism. The failure of one process does not impact the other functions in the router. The concept of containers fits exactly this description. Every process in the router is considered as one entity and comes with its own environment and libraries, fully isolated from the other processes. Each of these processes has a dedicated function within the router

  • It has built-in redundancy, which includes a monitoring component and a mechanism to automatically restart within a few milliseconds a process that stopped working

  • It allows for a complete disaggregation, keeping the forwarding plane functions within the hardware router, while the control plane could be hosted outside, on a secured cloud platform. From the NOS standpoint, it does not matter where the components are. As long as there is a reliable communication channel, the routing functions will properly do their job.

  • It can duplicate network functions easily as and when needed, to scale the performance based on the needs of the growing traffic.

  • It can upgrade some functions while leaving others untouched and do this while the router keeps forwarding traffic. No need to stop operations.

  • It can orchestrate all these functions, ensuring they act properly and timely.


Applying Cloud Technologies to Networking


Fortunately, the vision of a cloud-based NOS is no longer a dream. It already exists and applies to the world of disaggregated open networking. ExaNOS is a live example of it, powering internet peering networks to ensure high-performance connectivity to content and users.


ExaNOS is open by nature, it works on different hardware platforms from global manufacturers, such as Delta Networks, Ufispace and Edgecore Networks.


It is also open to third-party applications, which can influence routing decisions and retrieve much information from the flowing traffic. DDoS mitigation is a perfect case of third-party application, which can apply in real time thousands of filters to either reject or throttle any suspicious traffic. BGP Peering optimization is another use case of application to balance traffic across multiple peering links based on specific criteria; performance, link cost, least hops….


ExaNOS network functions are based on containers. Each function comes with its full environment and can work on a standalone basis. It allows for a system protection against faults and shields the router from a general failure. This system architecture enables a linear scalability, which is only constrained by the underlying hardware resources.


ExaNOS uses TIPC, a standard communication channel used within Linux Operating System to ensure a fast and reliable connection between network functions and the outside world.


TIPC has the flexibility to communicate within the router or outside of it, so much so that you could put the control plane on a secured cloud platform and keep the data plane within the router, without any significant change in the NOS.


Redundancy is also an important component of ExaNOS. Together with the hardware and the network protocols redundancy, the fault protection mechanisms within the OS offered by ExaNOS provide an overall carrier-grade availability.


While Cloud Computing has really flourished over the past 15 years, the networking industry still has a lot to catch up on, but the near future is bright, thanks to disaggregated open networking.



Comments


bottom of page