Data is Like Gold in Network AI

And not just any data but precision packet data.? This is the conclusion I came away with after an in-depth assessment of cPacket’s “AI for Network Observability” solution (https://www.cpacket.com).? During the last several years, there have been many vendors pitching that they have successfully applied AI/ML to network management, monitoring and observability, but only to fall short of expectations.? What these vendors got wrong was their presumption that the ML algorithms can find the answers, but we now know that the answer is in packet data and the algorithms are merely the means to find the answers.? The importance of packet data for AI is obvious if you can imagine what is happening inside the network wire which will show packets carrying vital content for you, me and end points in the digital economy.? Any other form of network data would be just an attempt to represent the packet behavior.? Packet data is the source of truth in network visibility.

cPacket is one of the few who understands the importance of packet data for AI, and who is delivering a viable solution with tangible customer values.? Coincidentally, the packet analytics industry has plenty of problems for AI to solve.? While packet is the tool of choice, it is complex to use and many organizations have not widely adopted it due to the need for expert network talents. The industry has been yearning for easier packet tools and this is where cPacket?is focused on solving.? cPacket employs an algorithm model with unsupervised learning, but where they stand out is the care and feeding surrounding their model.? Based on my analysis, there are three ingredients about cPacket’s implementation that make it compelling:? Precision packet data, Groomed datasets, and Training.

Precision Packet Data

In the packet analytics industry, cPacket has been known to be the best in collecting and processing packets with outputs in millisecond resolution and nanosecond accuracy.? This is the best in the industry and the secret is in the company’s use of purpose-built silicon to deliver the results at a line rate as high as 100 Gbps and above.? As an example, if you want to use ML to determine anomalies surrounding microburst congestion in data centers which occur in millisecond intervals, you will need a tool like cPacket to derive metrics at the same interval or shorter to detect the spiky traffic.? Such fine grain data is the first step in AI/ML development.

Groomed Datasets by Network Experts

To train an ML model, domain expertise is needed to condition the datasets with related features/metrics and to gather a high volume of diverse examples. ? cPacket started as a network silicon company and has built a core team of in-house network experts.? This knowledgebase is necessary to prepare the packet datasets for ML training.? Extracting and populating datasets with the right metadata and metrics based on network performance relationship is a crucial prerequisite to begin training.??

Training, Training, and Training

cPacket invested six years of R&D to train an ML model that is practical and has immediate time to value.? ML training is like practice drills in sports.? The more you do it, the more your muscles/AI can act instinctively.? Using data from several production networks, cPacket has been able to continuously train and evaluate their ML model to gain insights from oceans of precision packet data.? This resulted in an AI solution that can perform many of the tasks of a seasoned network guru.?

There are a lot of tasks and workflows in network observability that require deep packet expertise.? Examples of some common ones are isolating application latency across hybrid networks and multi-layer stack, SSL and TCP transaction glitches, calculating SYN to SYNACK ratio to detect DDOS, and detecting malicious vulnerability scanners.? Using cPacket’s AI for Network Observability, these tasks can now be accomplished in software and in a shorter time.? And the solution gets better the more one uses it:? it employs a form of reinforcement learning that adapts to the user context and feedback.? For many understaffed network teams, such a solution frees up days and weeks of valuable network expert time to perform more strategic network optimization.? Furthermore, the easy-button usability opens the door for more users (Application Owners, Level 2 NOC operators, Cloud architects) to adopt packet analytics, and neighboring tools (Network fault and performance tools, Application and server monitoring tools) to integrate its intelligence.? Now that we are in the era of AI, packet data and packet vendors like cPacket have the right mix to produce practical network AI solutions.