Machine Learning in Enterprise Networking Solutions
Cisco Systems, Inc.
The last few years have seen an explosion in the number of global companies that have started leveraging machine learning (ML) to convert data to business opportunities. Is there a role of machine learning in enterprise networking? The answer is a loud and resounding YES. However, more often than not, I see a number of IT companies starting this exploration of ML because it makes headlines and helps creates a buzz. This approach would fizzle out as it would not solve any major problems for their customers.
At Cisco, we take a different approach to this – we talk to our customers and partners, including our very own Cisco IT/InfoSec who serve as our internal customers. Given our huge customer base, we also look at several industry trends across multiple verticals. We use this approach to identify the right set of issues to focus on. Some of these problems warranted us to leverage machine learning to come up with the right solution. Once we identify the use case, we go about in a very systematic way to collect and curate the data to be able to identify the right set of features that would help us in achieving the desired outcomes. We then validate the algorithms by measuring the efficacy against the collected data.
There are a few key components that are required for the success of machine learning based solutions.
- The most important component required for the success of any machine learning based solution is the availability of data that can be used for training, and that is an area where Cisco has a significant advantage. We have by far the widest deployment base in the industry. This helps in ensuring that we build our learning based on robust data features which can be cross-validated against perhaps the world’s largest repository of network data.
- The second component is a platform that enables the collection and curation of data at scale. This component ensures the focus on the machine learning algorithms needed to be built to solve customer problems, rather than on the way the data is collected and shared. The Network Data Platform (NDP), which is part of Cisco’s recently launched DNA-Center, helps in the collection and correlation of data across the enterprise.
- The third component is network telemetry that manages a collection of the required data and export from the network elements. Some of the data features need to be made available to the analytics engines with minimal latency via streaming telemetry. Cisco's latest IOS-XE software stack supports streaming telemetry protocols.
Based on our conversations with customers, we strongly believe that there are several important problems that will require machine learning based solutions. One such problem we came across was the detection of malware in encrypted traffic without decryption. Cisco's recent launch of Encrypted Traffic Analytics addresses this problem. The network exports data which is leveraged by a cloud-based analytics engine to achieve the desired outcome. The cloud analytics engine has a global context which enables the collection of large volumes of data for training purposes, which in turn has a direct impact on the efficacy of the solution.
There are additional use cases in the context of security, IoT, and wireless that we're currently working on, which will be delivered via the DNA-Center. As mentioned above, the right way to do this is to gather large volumes of data for curation, which is precisely what we're doing now. We believe that these solutions will strongly establish the natural role that machine learning plays in enterprise networking solutions.
About the Author
Sarav Radhakrishnan is a Distinguished Engineer and a 17+ years Cisco veteran who has worked on a number of products and solutions during his tenure in Cisco.Sarav is the architecture lead for the highly profitable catalyst switching portfolio.
He has a proven track record of driving innovation in the portfolio and productizing the innovations. His current research and development interests include security, wireless, QoS, LiFi, virtualisation and machine learning.