Server for monitoring: Zabbix, Prometheus, Grafana

To create a reliable and scalable monitoring server capable of processing large volumes of data and ensuring high availability, a dedicated server with a powerful processor (4+ cores, 3+ GHz), sufficient RAM (16 GB for small installations, 64 GB for medium and large), and fast NVMe disks for the database is optimal, which is especially critical for systems like Zabbix or Prometheus with Grafana.

Which Dedicated Server to Choose for Monitoring: Zabbix or Prometheus + Grafana?

Choosing a platform for your monitoring server is one of the key decisions affecting the efficiency of your IT infrastructure. Zabbix and the Prometheus + Grafana stack are two of the most popular solutions, each with its own advantages and resource requirements. Zabbix is a comprehensive solution with agents, a server, and a web interface, storing data in a relational database. Prometheus is a system for collecting metrics with its own Time Series Database (TSDB) storage, while Grafana is a powerful tool for visualizing data from various sources, including Prometheus. Your choice will depend on the scale of your infrastructure, the type of metrics collected, architectural preferences, and available resources.

Zabbix Server: Resource Requirements and Scaling

The Zabbix server is the central component that collects data from agents, proxies, and other sources, processes it, executes triggers, and stores it in the database. The main resource consumers in Zabbix are:

• Zabbix Server Process: Processes incoming data, performs checks. Requires CPU and RAM.
• Database (MySQL/PostgreSQL): The most resource-intensive component. Stores all metrics, events, and history. Requires many I/O operations (especially writes), RAM for caching, and a powerful CPU.
• Zabbix Web Interface: A PHP application running on a web server (Apache/Nginx). Requires CPU and RAM, especially with active use.

Resource requirements for the Zabbix server depend significantly on the number of hosts, data items, collection frequency, data retention interval, and the number of users. Below are general recommendations:

• Small Installations (up to 100 hosts, 1000-2000 NVPS – New Values Per Second):
  • CPU: 2-4 vCPU (Intel Xeon E3/E5, AMD EPYC).
  • RAM: 8-16 GB (for Zabbix Server and DB).
  • Disk: 200-500 GB NVMe/SSD. NVMe is critical for the DB.
  • Retention: 7-30 days.
• Medium Installations (100-500 hosts, 2000-10000 NVPS):
  • CPU: 4-8 vCPU (Intel Xeon E5/E7, AMD EPYC).
  • RAM: 32-64 GB.
  • Disk: 1-2 TB NVMe. RAID10 is highly recommended for performance and fault tolerance.
  • Retention: 30-90 days.
• Large Installations (500+ hosts, 10000+ NVPS):
  • CPU: 8-16+ vCPU (Intel Xeon Gold/Platinum, AMD EPYC).
  • RAM: 64-128+ GB.
  • Disk: 2-4+ TB NVMe RAID10. May require splitting the DB across multiple disks or servers.
  • Retention: 90+ days.

For high-load systems, Zabbix can use proxy servers to distribute the data collection load, which helps offload the central Zabbix server.

Prometheus Hosting and Grafana Server: Features and Requirements

The Prometheus + Grafana stack offers a more decentralized approach to monitoring. Prometheus collects metrics, and Grafana visualizes them.

Prometheus Hosting

Prometheus operates on a "pull" model – it scrapes metrics from target systems over HTTP. It has its own Time Series Database (TSDB), optimized for storing time series data, which reduces its dependency on external DBMS compared to Zabbix.

Requirements for Prometheus hosting:

• CPU: Prometheus is not very CPU-intensive unless there are complex queries or a large number of recording rules. 2-4 vCPUs are usually sufficient for medium installations.
• RAM: Used for caching active time series. 8-32 GB RAM will be sufficient for most cases.
• Disk: The most critical resource. Prometheus writes data intensively. Fast disks are required, preferably NVMe, with a large capacity. Disk size depends on the number of metrics, their cardinality, and retention. For example, 100,000 active time series with a 30-day retention might occupy 500 GB - 1 TB. For a big data analytics server, where Prometheus is often used, NVMe disks are mandatory.

Example Prometheus configuration for data storage:

storage:
  tsdb:
    path: /prometheus
    retention.time: 30d
    wal-compression: true

Grafana Server

The Grafana server is a lightweight web application that connects to various data sources (Prometheus, Zabbix, InfluxDB, PostgreSQL, etc.) and visualizes them. Grafana itself does not store large volumes of metrics, only its configurations, dashboards, and users.

Grafana requirements:

• CPU: 2 vCPU.
• RAM: 4-8 GB.
• Disk: 50-100 GB SSD/NVMe (for OS and configurations).

Often, Prometheus and Grafana are deployed on a single monitoring server or in containers on a single dedicated server for Docker.

Comparing Zabbix and Prometheus+Grafana for a monitoring server

The choice between Zabbix and the Prometheus+Grafana stack for your monitoring server depends on the specific tasks and preferences.

Characteristic	Zabbix	Prometheus + Grafana
Architecture	Monolithic (server, agents, DB, web interface). Push and Pull models.	Decentralized (Prometheus - collection and storage, Grafana - visualization). Primarily Pull model.
Data Storage	Relational DBs (MySQL, PostgreSQL). Requires tuning and powerful disks.	Built-in Time Series Database (TSDB). Optimized for time series, very I/O intensive.
Alerting	Built-in, flexible, with many conditions and actions.	Prometheus Alertmanager. Powerful, but requires separate configuration.
Visualization	Built-in web interface, dashboards. Functional, but less flexible than Grafana.	Grafana. Leader in visualization, many data sources, flexible dashboards.
Discovery	Low-Level Discovery (LLD).	Service Discovery (Kubernetes, Consul, etc.).
Scaling	Horizontal (via proxies) and vertical (more powerful DB server).	Horizontal (via Federation, remote storage, sharding) and vertical.
Disk I/O Requirements	High for DB (read/write). NVMe is critical.	Very high for Prometheus TSDB (intensive writes). NVMe is mandatory.
Configuration Complexity	Simpler initial setup, but more complex DB tuning.	More modular, requires configuring several components, but more flexible.
Usage	Traditional monitoring of servers, network equipment, applications.	Monitoring of cloud environments, microservices, containers (Kubernetes), dynamic infrastructures.

How to Choose a Dedicated Server for Zabbix, Prometheus, or Grafana?

Choosing a dedicated server for monitoring directly impacts your system's performance and stability. The advantages of a dedicated server over a VPS are obvious: guaranteed resources, no "noisy neighbors," and full control over the hardware. This is especially important for I/O and stability-critical systems like a Zabbix server or Prometheus hosting. Cloud vs Dedicated: when the cloud is not needed, shows that for such workloads, a dedicated server often proves to be more cost-effective and performant. When choosing a server, focus on the following parameters:

1. Processor (CPU): For Zabbix, clock speed and core count are important (for processing triggers and DB queries). For Prometheus – rather the number of cores for parallel processing of queries and metric collection. Look for Intel Xeon E5/E7/Gold or AMD EPYC with a high clock speed (3.0+ GHz) and 4+ cores.
2. Random Access Memory (RAM): The more, the better. Zabbix and its DB actively use RAM for caching. Prometheus also benefits from a large amount of RAM for its TSDB. A minimum of 16 GB, optimally 32-64 GB for medium installations.
3. Disk Subsystem: Critically important parameter.
   • Disk Type: Only NVMe SSD. SATA SSD may be sufficient for the OS and Grafana, but for Zabbix databases and Prometheus TSDB, NVMe provides the necessary I/O speed.
   • Capacity: Depends on the volume of collected data and its retention period. For Zabbix with 90-day retention on 500 hosts, 1-2 TB NVMe will be required. For Prometheus with the same retention – similarly.
   • RAID: For fault tolerance and increased performance (especially writes), RAID10 from NVMe disks is recommended.
4. Network Interface Card (NIC): 1 Gbps Ethernet minimum, 10 Gbps for large installations with many agents or for centralized data collection from multiple locations.

Example Dedicated Server Configurations for Monitoring

Valebyte offers various dedicated server configurations for monitoring that are suitable for Zabbix, Prometheus, or Grafana.

Installation Type	Processor	RAM	Disk	Network Card	Approx. Cost/Month
Small (up to 100 hosts, Zabbix/Prometheus+Grafana)	Intel Xeon E3-12xx / E5-26xx (4C/8T, 3.2+ GHz)	16-32 GB DDR4	500 GB NVMe RAID1	1 Gbps	From $80
Medium (100-500 hosts, Zabbix/Prometheus+Grafana)	Intel Xeon E5-26xx / Gold (8C/16T, 2.8+ GHz)	32-64 GB DDR4	1-2 TB NVMe RAID10	1 Gbps	From $150
Large (500+ hosts, Zabbix/Prometheus+Grafana)	Intel Xeon Gold/Platinum / AMD EPYC (12C/24T+, 2.5+ GHz)	64-128 GB DDR4	2-4 TB NVMe RAID10	10 Gbps	From $250

These configurations provide a reliable foundation for your monitoring server. For very large installations, clustering or distributed solutions may be required.

Optimizing Monitoring Server Performance: Valebyte Tips

After choosing a suitable dedicated server for monitoring, it is important to focus on software optimization.

1. Database Tuning (for Zabbix):
   • Use MySQL InnoDB or PostgreSQL.
   • Configure caching parameters: innodb_buffer_pool_size (MySQL) or shared_buffers, work_mem (PostgreSQL) up to 70-80% of available RAM.
   • Regularly clean old data (housekeeping in Zabbix).
   • Index tables.
   Example MySQL configuration in my.cnf:

   [mysqld]
   innodb_buffer_pool_size = 70%_OF_RAM
   innodb_log_file_size = 256M
   innodb_flush_log_at_trx_commit = 2
   max_connections = 500

2. Prometheus TSDB Optimization:
   • Set an adequate retention.time to save disk space.
   • Use wal-compression: true to reduce I/O.
   • Limit metric cardinality (the number of unique label combinations).
3. Zabbix Server Configuration:
   • Increase the number of StartPollers, StartDiscoverers, StartHTTPPollers, and other Zabbix Server processes depending on the load.
   • Place the Zabbix Server and database on a single server with fast NVMe disks or separate them onto different servers for maximum performance.
4. Using Proxies (for Zabbix): For distributed environments or a large number of hosts, use Zabbix Proxy to reduce the load on the central Zabbix server and decrease network traffic.
5. Monitoring the Monitoring Itself: Set up performance monitoring for the monitoring server itself (CPU, RAM, Disk I/O, network traffic). This will allow timely identification of bottlenecks.
6. Operating System Choice: Linux (Ubuntu Server, CentOS, Debian) is the standard for such solutions.
7. Cost Reduction: Regularly review your data retention policy and delete unnecessary metrics to avoid overspending on disk and memory resources. This is one way to reduce server infrastructure costs.

Conclusion

Choosing the optimal monitoring server is an investment in the stability and performance of your IT infrastructure. For most environments requiring comprehensive monitoring and detailed configuration capabilities, a Zabbix server on a dedicated server with NVMe disks will be an excellent solution. If your infrastructure is dynamic, built on microservices or Kubernetes, and you value visualization flexibility, then the Prometheus hosting and Grafana server stack on a dedicated server with powerful disk I/O will be the preferred choice. Valebyte offers dedicated servers ideally suited for both platforms, ensuring the necessary performance and reliability.