# Celestia light node performance analysis

By [FTP Crypto](https://paragraph.com/@sooqa) · 2023-04-14

---

Conducted a performance analysis of my light node. For that purpose I was using scripts from celestia team

[https://github.com/P-OPSTeam/celestia-tools/tree/main/grafana](https://github.com/P-OPSTeam/celestia-tools/tree/main/grafana)

With the help of those scripts I ve properly set up prometheus and grafana interaction.I also added some metrics that seemed useful to me:

*   celestia\_das\_busy\_workers\_amount
    
*   celestia\_das\_sample\_time\_hist\_count
    

You can check all the _dashboards_ I used via the link

[http://135.181.33.247:3000/d/sAAM7CLVk/processes-monitoring?search=open](http://135.181.33.247:3000/d/sAAM7CLVk/processes-monitoring?search=open)

So I ve analysed the data coming from node and noticed the following:

### Sampled headers

More or less equal number of sampled headers is being generated every hour

![](https://storage.googleapis.com/papyrus_images/f40646831286817a8222be46cb9e437427c5fc864dc40754b2a1c28ea14c86b0.png)

So every minute from 4 to 6 sampled headers are being generated

[https://docs.google.com/spreadsheets/d/1nYkl13oLsLcqzEit9lNYlTMDJV9j0K00-uD9galbS\_w/edit?usp=sharing](https://docs.google.com/spreadsheets/d/1nYkl13oLsLcqzEit9lNYlTMDJV9j0K00-uD9galbS_w/edit?usp=sharing)

In total for the last 27 hours 8735 sampled headers were generated. This number can be divided into:

*   width 2 headers (59)
    
*   width 4 headers (6739)
    
*   width 8 headers (1938)
    

Headers with width 4 appeared to be the most common

### Head syncing

Head also grows by 4-6 blocks every minute (most commonly by 5-6)

![](https://storage.googleapis.com/papyrus_images/7ff8512ebd7395d7745047e62ce939c9a2c8c2e96a1e425f11bba6b603b92097.png)

### DAS busy workers

Noticed an interesting pattern here

![](https://storage.googleapis.com/papyrus_images/f56b1f6d68f82d22c070cf335ae0447634c7a19166eedf5962d7299da9b7cb89.png)

So in my case two workers get busy once in every 1.5-2 hours. One workers is usually busy for no more than one minute.

### System load

Some metrics here can vary, however in my case light node takes 1.86 GB of resident memory, 9.76 GB of virtual memory

![](https://storage.googleapis.com/papyrus_images/eecf47417909e485e40a28122ea07e98048fb08c30ef26e4df86fa4df9cc9d9e.png)

CPU performance of celestia tasks in my case looks like:

_Min_ 0.0431 ms _Max_ 0.07 ms _Avg_ 0.0433 ms

![](https://storage.googleapis.com/papyrus_images/0c7621f9e6a96eac507dc26c286481b3e401d24dd51f1230c79f585cc29e5e34.png)

Based on 1 hour data set from the google doc above

> _Analysis conducted by Danylo Soloviov Discord: Daniel\_#6773_

---

*Originally published on [FTP Crypto](https://paragraph.com/@sooqa/celestia-light-node-performance-analysis)*