Network Flow Monitor
Near real-time network performance monitoring for your AWS workloads — detect issues, attribute causes, and reduce troubleshooting time.
What is Network Flow Monitor? ▾
Network Flow Monitor is a feature of Amazon CloudWatch Network Monitoring that uses fully-managed, lightweight agents to provide near real-time visibility into network performance across your AWS compute resources, services, and infrastructure.
It analyzes actual TCP traffic between your workloads — EC2 instances, EKS pods, and AWS services like S3 and DynamoDB — and surfaces performance and availability metrics without requiring active synthetic probes.
Problems It Addresses ▾
| # | Problem | Impact |
|---|---|---|
| 1 | Limited visibility into AWS network performance | Traditional monitoring tools cannot see inside AWS-managed network infrastructure. When latency spikes occur, teams lack data to determine if the problem is their application, a security group, or the AWS backbone itself. |
| 2 | Long MTTD / MTTR | Without real-time network telemetry, teams spend hours running synthetic tests and reviewing logs to isolate network issues — delaying incident resolution and impacting SLAs. |
| 3 | Inability to attribute cause — app vs. infrastructure | When performance degrades, engineers cannot quickly determine whether the root cause is in their code, the network path, or the AWS infrastructure. |
| 4 | Hidden inter-AZ and NAT gateway costs in EKS | Microservices on EKS generate cross-AZ traffic ($0.02/GB roundtrip) and NAT gateway processing charges ($0.045/GB) that are invisible at the pod level. |
| 5 | No pod-level network observability in Kubernetes | Platform teams cannot see which pods are "top talkers," which services generate the most east-west traffic, or how traffic flows to external endpoints. |
How Network Flow Monitor Solves These Issues ▾
| Solution | Description |
|---|---|
| Passive, real-time monitoring | Lightweight eBPF-based agents analyze real TCP traffic at the kernel level — no synthetic probes or packet captures needed. Metrics are available in near real-time without impacting application performance. |
| Network Health Indicator (NHI) | NHI automatically determines whether degradation is caused by AWS infrastructure or your workload. A value of "Degraded" means AWS network issues were detected — allowing teams to immediately focus on the right layer. |
| Network path topology visualization | When degradation occurs, Network Flow Monitor displays all components in the network path with service icons and resource IDs, pinpointing exactly where the bottleneck is. |
| Container Network Observability for EKS | Provides pod-level visibility with a Service Map, Flow Table (top talkers by AWS service, cluster, or external traffic), and Performance Metrics (ingress/egress, packet counts, bandwidth). |
| Cost attribution with inter-AZ and NAT tracking | Identify exactly which pods and services generate cross-AZ traffic and NAT gateway charges. Enables targeted optimizations like Traffic Distribution Control and VPC endpoints. |
| Workload Insights | Top contributor analysis surfaces the highest-impact network flows automatically. Engineers can spot emerging issues before they become incidents and create targeted monitors. |
Supported Platforms ▾
The Network Flow Monitor agent can be installed on the following compute platforms:
| Platform | Agent Type | Installation Method | Status |
|---|---|---|---|
| Amazon EC2 (Linux) | Standalone agent | AWS Systems Manager Distributor — AmazonCloudWatchNetworkFlowMonitorAgent | GA |
| Amazon EKS | EKS Add-on (eBPF DaemonSet) | EKS Console → Observability → Network Monitoring or via EKS API/CLI add-on | GA |
| Self-managed Kubernetes | Kubernetes DaemonSet | Manual deployment via manifests | Supported |
CloudWatchNetworkFlowMonitorAgentPublishPolicy IAM policy attached to their role.
EKS Container Network Observability Features
- Service Map — Dynamic visualization of inter-pod communication. See your application topology at a glance, with retransmission and data transfer metrics per flow.
- Flow Table (3 views) — AWS Service view: traffic to S3, DynamoDB, etc. Cluster view: east-west pod traffic. External view: traffic leaving AWS (NAT gateway).
- Performance Metrics Endpoint — Scrape OpenMetrics-compatible data (ingress/egress flows, packet counts, bytes, allowance-exceeded counters). Export to Prometheus and Grafana.
How It Works ▾
Architecture Overview
(eBPF on EC2/EKS)
(Aggregation & NHI)
(Workload Insights & Monitors)
bpf_sock_ops structure from the Linux kernel — no payload access.Step 1: Enable Network Flow Monitor
Navigate to CloudWatch → Network Monitoring → Flow Monitors. Enable the feature for your account in the target region. This creates the monitoring scope and configures service-linked roles. (One-time per region, ~30 min setup.)
Step 2: Install Agents
Deploy the lightweight agent on your compute resources. For EC2: use Systems Manager Distributor (AmazonCloudWatchNetworkFlowMonitorAgent). For EKS: enable Container Network Observability via the EKS console or install the AWS Network Flow Monitor Agent add-on.
Step 3: Agent Collects Metrics
Agents passively analyze TCP connections and collect: RTT, retransmissions, retransmission timeouts, and bytes transferred. They access only the bpf_sock_ops structure — local/remote IPs, ports, and TCP counters. No payload data is accessed.
Step 4: Metrics Aggregated & Published
Performance metrics are sent to the NFM backend, which aggregates them by subnet, Availability Zone, VPC, and AWS service. Metrics are available in the AWS/NetworkFlowMonitor CloudWatch namespace and surfaced in the console.
Step 5: Review Workload Insights
The Workload Insights tab shows top contributors by metric type across all monitored flows. Identify impairments, review trends, and determine which flows need dedicated monitors.
Step 6: Create Flow Monitors
Create targeted monitors by selecting local resources (subnet/VPC/AZ) and remote resources (specific subnets, services, or "Everywhere"). Monitors provide detailed performance data, network path topology, and the Network Health Indicator (NHI) for root cause attribution.
list-monitors, start-query-monitor-top-contributors, and get-query-results-monitor-top-contributors allow automation of traffic analysis and alerting.
Key Metrics ▾
Network Flow Monitor collects the following TCP-level performance metrics passively from actual traffic:
All metrics are available in the AWS/NetworkFlowMonitor CloudWatch namespace and are aggregated by subnet, Availability Zone, VPC, and AWS service.