System requirements

Reference
2-min read
Published Nov 19, 2024

This page describes system requirements for the Dynatrace OpenTelemetry Collector Distribution for different use cases.

Hardware used

The numbers below were gathered using an Azure virtual machine of type Dadsv5-series (AMD EPYC 7763v Genoa CPU) with 4 vCPUs and 16GB RAM.

CPU and memory requirements for traces, metrics and logs (combined)

The requirements for the Dynatrace Collector are based on a load scenario with the following numbers per second:

1000 traces (~1.2 KB each)
1000 metrics (~1.2 KB each)
1 MB logs

The recommended resources for this combined scenario are:

500 MiB RAM
500 mCPU

If you need additional data processing (for example, filter or transform processors), system requirements will increase.

Check out specific scaling and infrastructure architecture considerations on our separate page.

Specific load scenarios

The Dynatrace Collector was also tested under heavier scenarios than above but only with single signal types. The following table shows performance data based on the following base data sizes:

Traces: ~1 KiB per trace
Metrics: ~3 KiB per metric
Logs: ~3 KiB per line

This table shows throughput scenarios based on the above data sizes with different amounts of load and different used protocols.

Scenario (traces, metrics, logs per second)

CPU cores

RAM (MiB)

OTLP-HTTP 10k traces

0.25

100

OTLP-HTTP 100k traces

1.5

120

OTLP-HTTP 10k metrics

0.25

110

OTLP-HTTP 100k metrics

100

Syslog 10k logs 1 per batch

0.2

100

Syslog 10k logs 100 per batch

0.2

100

Syslog 70k logs 1 per batch

100

Syslog 70k logs 100 per batch

0.5

110

Prometheus scrape performance

Additional metrics based load scenarios were done based on Prometheus scraping with the following base settings:

1s scrape interval
Metrics data size: ~3 KiB per metric

This table shows the load results with different scenarios.

Scenario

CPU cores

RAM (MiB)

1 endpoint (10k data points each)

0.5

300

1 endpoint (1k data points each)

0.1

140

5 endpoints (1k data points each)

1.5

250

10 endpoints (1k data points each)

500