🌌 IPFS Measurement Report - 2023 Week 41 - 2023-10-16

We have created a new IPFS measurement report for calendar week 38 in 2023 from 2023-10-09 to 2023-10-16. You can find it here:

Some observations from the report and the results at probelab.io:

  • Around 7k nodes left the network from Friday to Saturday. Also clearly seen at: IPFS DHT | ProbeLab
    – Most of them seem to have been Kubo 0.18 nodes
    – @dennis-tra how come we do not see the decline in the number of DHT Server nodes at: IPFS DHT | ProbeLab? That’s weird, no?
  • The Median DHT Lookup Performance over time is going back down after the quick bump in the last couple of weeks: IPFS DHT | ProbeLab
  • On the contrary the Median DHT Publish Performance over time is going back up after a performance increase the last couple of weeks: IPFS DHT | ProbeLab
  • Not sure we have clear explanation about any of these two events @dennis-tra ?
  • There’s a decline in the number of “Uncategorised” nodes in the DHT Server Availability, classified by region plot: IPFS DHT | ProbeLab. Probably now showing under “North America”?
  • About 8k nodes moved from “Online” to “Mostly Online” in the DHT Server Availability, classified over time plot at: IPFS DHT | ProbeLab. I’m wondering if they’ll get back to “Online” eventually.

Because the graph on probelab.io shows the number of unique server peer IDs over the course of one week. Because these 7k peers left within the last week I expect a decline of ~15% next week.

Yeah… my suspicion is that it’s due to a redeployment although I can’t explain the dynamic here. Could it have something to do with a cold vs. warm routing table? I would have expected that this effect would go away after a few hours and wouldn’t last a few days. I’ll check the data later :+1:

The improvement in publication performance also coincides with the redeployment. The decrease in performance coincides with the 7k peers leaving the network. If it’s just correlation or already causation I don’t know at the moment.

This is due to the 7k peers that have left the network. They left the network from Friday to Saturday night, so they were online for ~5/7 → ~71% of the week. Due to our classification that peers with 40% to 80% availability are “mostly online”, these 7k peers that have left the network fall into this category.

1 Like