@cryptotaxi247 / netdata-1
Code Commits Issues Pulls Sessions CI/CD
master
master
md 122 lines 5.6 KB
Rendered Raw
1 # Agent Performance Optimization Guide
2
3 While Netdata Agents work seamlessly out-of-the-box with comprehensive monitoring, you can tune their configuration for better performance when needed.
4
5 ## Why optimize your Agent
6
7 By default, your Netdata Agent provides:
8
9 - **Automatic Application Discovery**: Continuously detects and monitors applications on your node
10 - **Real-time Metric Collection**: Collects metrics every second
11 - **Health Monitoring**: Actively tracks health status with built-in alerting
12 - **Machine Learning**: Trains anomaly detection models for each metric ([Anomaly Detection](/src/ml/README.md))
13
14 These features deliver comprehensive monitoring but consume system resources. You might need to optimize when running Agents on resource-constrained systems or when scaling your monitoring infrastructure.
15
16 :::note
17
18 See [Resource Utilization](/docs/netdata-agent/sizing-netdata-agents/README.md) for detailed Agent resource requirements.
19
20 :::
21
22 ## How to optimize performance
23
24 Here's how each optimization strategy reduces resource usage:
25
26 | Optimization Strategy | Reduces CPU | Reduces RAM | Reduces Disk IO |
27 |-----------------------------------------------------------------------|-------------|-------------|-----------------|
28 | [Set up Parent-Child architecture](#set-up-parent-child-architecture) | ✓ | ✓ | ✓ |
29 | [Disable unneeded collectors](#disable-unneeded-collectors) | ✓ | ✓ | ✓ |
30 | [Reduce collection frequency](#reduce-collection-frequency) | ✓ | | ✓ |
31 | [Adjust metric retention](#adjust-metric-retention) | | ✓ | ✓ |
32 | [Switch to RAM mode](#switch-to-ram-mode) | | ✓ | ✓ |
33 | [Turn off ML on Children](#turn-off-ml-on-children) | ✓ | | |
34
35 ## Set up Parent-Child architecture
36
37 Transform your monitoring by using Parent nodes as centralization points. Parents collect and aggregate data from multiple Child nodes, significantly reducing the load on individual systems.
38
39 In this setup:
40
41 - **Children** stream their metrics to Parents instead of storing everything locally
42 - **Parents** handle data aggregation, storage, and dashboard queries
43 - **You** access all metrics through the Parent nodes
44
45 :::tip
46 This architecture works especially well in production environments where you monitor many systems. Learn more in our [Centralization Points documentation](/docs/deployment-guides/deployment-with-centralization-points.md).
47 :::
48
49 ## Disable unneeded collectors
50
51 Reduce resource usage by turning off [Plugins or Collectors](/src/collectors/README.md) you don't need.
52
53 :::warning Important
54 Only active collectors consume resources. Inactive plugins and collectors shut down automatically, so you only save resources by disabling those currently running and collecting metrics.
55 :::
56
57 Follow our [configuration guide](/src/collectors/REFERENCE.md) to identify and disable specific collectors.
58
59 ## Reduce collection frequency
60
61 Save CPU and disk IO by collecting metrics less frequently. If you don't need per-second precision, or if your Agent consumes too much CPU during periods of low dashboard activity, increase the collection interval.
62
63 This change:
64
65 - Significantly reduces CPU usage
66 - Decreases disk write operations
67 - Maintains meaningful monitoring capabilities
68
69 Learn how to adjust collection frequency in our [configuration guide](/src/collectors/REFERENCE.md).
70
71 ## Adjust metric retention
72
73 Control memory and disk usage by changing how long your Agent stores historical data. Shorter retention periods mean:
74
75 - Less RAM needed for in-memory metrics
76 - Reduced disk space requirements
77 - Faster Agent startup times
78
79 Configure retention settings using our [database configuration guide](/src/database/CONFIGURATION.md).
80
81 ## Switch to RAM mode
82
83 For IoT devices and Child nodes in [Parent-Child setups](/docs/deployment-guides/deployment-with-centralization-points.md), switch to RAM mode to eliminate disk operations entirely. This mode:
84
85 - Stores all metrics in memory only
86 - Eliminates disk IO for metric storage
87 - Significantly reduces overall resource usage
88
89 :::tip
90
91 Since Child nodes stream metrics to Parents, they don't need persistent local storage. RAM mode is ideal for this use case.
92
93 :::
94
95 Set up RAM mode following our [database configuration guide](/src/database/CONFIGURATION.md).
96
97 ## Turn off ML on Children
98
99 Optimize resource allocation by running Machine Learning only where it matters most. We recommend:
100
101 - **Enable ML on Parents**: They have the complete data picture and typically more resources
102 - **Disable ML on Children**: They focus on collecting and streaming metrics
103
104 To disable ML, edit your configuration using [`edit-config`](/docs/netdata-agent/configuration/README.md#edit-configuration-files):
105
106 ```text
107 [ml]
108 enabled = no
109 ```
110
111 :::tip
112
113 This configuration particularly benefits Child nodes, allowing them to focus on their primary role of collecting and streaming metrics to Parent nodes where ML analysis happens centrally.
114
115 :::
116
117 ## Next steps
118
119 1. **Identify your needs**: Determine whether you need optimization for resource constraints or architectural efficiency
120 2. **Start with architecture**: If monitoring multiple systems, implement Parent-Child setup first
121 3. **Fine-tune individual Agents**: Apply specific optimizations based on each system's role and resources
122 4. **Monitor the impact**: Use Netdata dashboards to confirm your optimizations haven't compromised monitoring visibility