Networking industry has been far too reserved. While technologies such as software-defined networking and Network Functions Virtualization are crucial elements of the network and have been in the industry well for the last half decade, what we needed is a more complete networking vision. For instance, what if we could apply recent advances in AI technologies to build immensely improved network? Especially, what if we can apply the concept of self-driving cars to networking? We want to make self-driving networks that can monitor, configure and maintain themselves as well as adapt to their operating domain with little or no human intervention at all.
The Self-Driving Network, like a self-driving car, is an autonomous network. It is predictive and adaptive to its own environment. The biggest advantage of this type of network is that it increases economies of scale and efficiencies as the same time decreasing operating costs. So, The Self-Driving Network delivers an optimized and customized quality of experience. Also, it is inexpensive to the end-user. The aim of this Network is to eliminate the “manual work” needed to keep the networks running.
What if we could apply recent advances in AI technologies to build an immensely improved network?
The concept of automation for e.g. self-equipping, self-diagnosing, and self-recovery, has been around for some time. But now with advances in AI and cloud techs, such fanciful believes are quickly becoming realities. AI enables the next level of networking.
Why Are We Not There with Networks?
Now, I am going to discuss about the obstacles the network industry may face along the way to his path to Self-Driving Networks. We must be aware of and prepared for the widely applicable effects of the Tech-disruption, in addition to the broader impacts of the Self-Driving Networks. Last year, Juniper Networks Inc published a paper called White Paper in which they talked about The Self-Driving Network Part 1: A Bolder Vision for the Industry. Let’s discuss about some of their points here and why it makes sense.
If we analyze the evolution of the rise of the self-driving car, we can come up with these three influencers that reveals why we are not there yet, and the way autonomous networks may evolve.
Finances
Let’s think about the reasons what was the economic reason behind the evolution of self-driving cars. The car making companies have been improving quality of cars for decades and prices continue to drop. But they never thought of self-driving cars.
However, looking at it from a different perspective, Google had a different economic motivation behind revolutionary work in self-driving car development. One reason would be, if they can release everyone from an hour of go back and forth every day, that additional time people can spend on the Internet, which in the end could create revenue for Google.
But the more convincing economic benefits lie with companies such as Uber. Because with Uber, drivers receive about 75% of each fare and if they can operate self-driving cars, then there will be reduction in the cost of goods sold (COGS) by 75%. As a result, the ride sharing companies are taking more interest in autonomous vehicle.
Let’s talk about networks now and we can easily understand why there was no such force behind self-driving networks.
Know-how
Another powerful factor was technology behind the revolution of self-driving cars. Google disrupted the auto-industry because it wanted to monetize the big data they have. It was just another project that was a part of their mission of monetizing the big data they gathered in last decade.
Interestingly, the self-driving vehicles emerged out of the coming together of three other unrelated technology mega-trends—cloud computing, mobile networks, and big data analytics.
Till now, there were few companies who talked about following the same self-driving cars evolution path and disrupting the network industry, but no one fully succeeded in doing so.
People
Forbes estimated the benefits to the society of self-driving cars could be as much as $642 billion per year in the U.S. alone based on calculations for injuries, lives saved, and time saved. So, the humanity can get a huge benefit if the self-driving cars come into the market and it’s going to sale and make a good return for the companies. There is no such type of advantage for self-driving networks.
Self-Driving Networks Timeline
1960s—2000: Packet-switching developed
Protocols for computer-to-computer communication
Algorithms for routing, switching, load balancing, congestion control
2005—now: Software-defined Networking (SDN)
Programmability and flexibility
Now—Self-Driving Networks
Autonomy: means Network intelligences, monitors and controls itself
Interactivity: Infra should be transparent and fun to interact with, especially for 3rd party users
4 Important Problems Ahead of Self-Driving Networks
Researchers and executive teams from the companies who are working on the self-driving networks, believes these 4 important problems must be solved for networking to advance.
Growing performance as Moore’s Law is slowing
We are improving the performances of new devices every year and Moore’s Law is slowing. All these new devices will require to grow by scaling out means they are becoming massively parallel instead of scaling up. So that is creating interconnected data centers with many thousands of servers. These servers are required to handle applications such as augmented VR, IoT, machine learning and many more.
Right now, too much manual efforts invested in the maintenance of existing networks
The previous problem is the cause of this problem. The thousands of servers will drive the formation of networks with thousands of nodes and those will be too complex to be functioned and maintained by humans and it’ll be difficult for humans to type at the command line interface. Automation will be vital to remove continued human interference to one in which network systems function autonomously.
Compound networks with many layers
Networks are built in orders that operate independently, producing substantial inefficiencies. This multi-layer method to manage IP and optical transport networks is not maintainable and that is why automated and harmonized organization layers must be advanced to reduce complexity as well as reducing the need for human interference.
Trusting on perimeter security.
We are relying on perimeter security now. That means our internal data assets can be protected from threats coming from public networks and third-party networks.
For automation, we are going to need strong authentication for sharing security tokens and creating trusted machine-to-machine communications.
Ongoing Progress for Self-Driving networks
Stanford Lab experiment
What does Self-Driving mean?
In the paper published by Stanford university, Prof. Balaji Prabhakar and Prof. Mendel Rosenblum talks about the self-driving networks.
They talk about a scenario in which we’re given a DCN and a workload or network jobs that arrive over time. We need to allocate resources like network, CPU, machine memory, and storage in a way so that the jobs which are arrived over time, can be processed quickly that means small job completion time. Also, in this scenario we can utilize the resources very efficiently.
NOW, the question is what is DCN?
A DCN that is a dynamic circuit network which is an advanced computer networking technology that combines packet-switched communication based on the IP with circuit-switched technologies.
DCN allows fixed bandwidths between different communication partners to be assigned for a short period of time.
The routers to be connected require special software to create DCN connections. We can compare the DCN connections with connections in a circuit-switched network for e.g. telephone network in which date can be transmitted from one subscriber to the other subscriber without being throttled by the actions of other subscribers. Also, very high bandwidths can be achieved in a DCN. So, we can transmit a large amount of data in a continuous data stream without any loss of data and we can transmit the data extremely quickly.
Experiment: Sense DCN from the edge
The team of Stanford Labs used NIC-based telemetry (for e.g. https://www.solarflare.com/packet-telemetry) to sense the DCN from the edge and recorded the results from the test. They used NIC-based telemetry because they are more scalable means edge observations are adequate data statistics. They used NICs which can be used for time-stamping probes/packets.
Platforms and Testbeds
Google testbed with 40G links, 5-stage Clos switching, Cisco 2960 switch and Stanford testbed with 1G links, 2-stage Clos switching
Algorithm
Input: 5-tuple flow IDs for network paths and Rx and Tx timestamps of probes
Equation for each packet:
Combine all packets: D= AQ + N
They solved the sizes of queue using Lasso Algorithm
These graphs were recorded as results.
Roadmap and progress
Their current focus was to sense at the edge and reconstruct network details. Also, collect Machine learning data, implementing network algorithms and pattern recognition which is learning network load from patterns like packet traces and memory utilization patterns.
Their Future work will focus on learning the best real-time responses Reinforcement Learning and integrating network controllers for real-time self-directed control.
Summary from Stanford Lab experiment
They believe that Self-Driving Networks is a project of multi-years. Their current model already has clock synchronization and network reconstruction.
They feel they are ready for broader deployment in which they are not just going to use telemetry but regressions, planning, policy setting etc.
Their focus for this project was learning but in next phase they’re going to include control as well.
Cisco Systems vs Juniper Networks
These both companies imagine a future of self-driving networks. Both differ on how to get there, but the both believe in intent-based networking which is a step toward creating a self-sufficient infrastructure, which highlights cognitive thinking with the use of AI and machine learning which also has the capacity to proactively sense and repairs network and security events.
Cisco data center operators can begin implementing intent-based networking with three products:
- Application Centric Infrastructure (ACI) is a data center which delivers intent over policy-based automation;
- Tetration Platform is a platform that analyzes the network in real-time and senses network and security glitches
- Network Assurance Engine (NAE) which is a software that confirms network behavior and guarantees that the network follows policies and can take counteractive action.
Also, Juniper released its Contrail Enterprise Multi Cloud software in April 2018.
That is an SDN controller that supports as the groundwork for its intent-based networking approach for the data center.
Juniper’s acquired cloud processes management company, AppFormix last year and united its AppFormix monitoring and intent-based analytics tools into this software. That allows IT managers to manage and monitor assignment policies from just a single command center.
Juniper Bots
Contrail TestBot, which allows operators to test network changes before they are applied AppFormix HealthBot, which uses machine learning to analyze network health and provide suggestions for improvements; Contrail PeerBot, which automates network peering.
Intent Based Networking
Both companies imagine a future of self-driving networks Both believe in intent-based networking. It is a step toward creating a self-sufficient infrastructure, which highlights cognitive thinking with the use of AI and machine learning which also has the capacity to proactively sense and repairs network and security events.
References
Wexler, S. (2017, April 20). The Self-Driving Network: SDN On Steroids. Retrieved from https://www.cio.com/article/3190556/networking/the-self-driving-network-sdn-on-steroids.html
Juniper Staff, (2017, June). The Self-Driving Network https://www.juniper.net/assets/us/en/local/pdf/whitepapers/2000656-en.pdf
Balaji Prabhakar and Mendel Rosenblum, (2017, April). Self-Driving Network https://platformlab.stanford.edu/pdf/Balaji.pdf
Dynamic circuit network. (2017, September 15). Retrieved from https://en.wikipedia.org/wiki/Dynamic_circuit_network
Wong, W. (2018, September 17). Intent-Based Networking in the Data Center Cisco vs. Juniper Retrieved from https://www.datacenterknowledge.com/networks/intent-based-networking-data-center-cisco-vs-juniper
Juniper envisages self-driving networks. (n.d.). Retrieved from https://www.computerworld.com.au/article/612449/juniper-envisages-self-driving-networks/
DCN. (2018, April 18). Retrieved from https://www.nfon.com/en_de/cloud-telephone-system/resources/glossary/dcn/
Artificial Intelligence networks Self-Driving
