How it works - Amazon CloudWatch

How it works

This section provides information about several aspects of how Network Flow Monitor works.

How Network Flow Monitor agents gather statistics

Agents in Network Flow Monitor are installed on AWS compute resources (Amazon EC2 and Amazon EKS), where they gather performance metrics and send them to the Network Flow Monitor backend. Agents do not have access to the payload of your TCP connections. Agents receive only what is called the "bpf_sock_ops" structure from the Linux kernel. This structure provides the local and remote IP address and the source and destination TCP port, as well as counters and round-trip times. For list of the TCP statistics collected and published by the agent, see View Network Flow Monitor metrics in CloudWatch.

The agent uses the Network Flow Monitor Publish API to send metrics to the Network Flow Monitor backend server.

Note: Network Flow Monitor supports up to approximately 5 million flows per minute. This is approximately 5,000 instances (Amazon EC2 and Amazon EKS nodes) with NFM agent installed. Installing agents on more than 5000 instances may affect monitoring performance until additional capacity is available.

How network flows are categorized in Network Flow Monitor

Network Flow Monitor categorizes network flows into classifications depending on where the flows originate and terminate.