As promised in our recent post at: ProbeLab’s Updated Performance Metrics , here’s a summary of the most important results from the lenses of ProbeLab’s monitoring tools.
This is the first report after quite a while, so will be a little more comprehensive, covering several aspects and overall performance. Follow up updates will focus primarily on “what has changed since the previous report”.
The intention of these reports is to direct efforts to parts of the protocol that need attention. Here’s the highlights:
-
Sharp drop in the number of DHT server nodes lately: https://probelab.io/ipfs/#chart-ipfs-servers-vs-clients-ts
-
IPFS Bootstrappers generally very stable - haven’t spotted any outage for a long while: https://probelab.io/ipfs/bootstrappers/
-
Kubo continues to be the dominant IPFS client: https://probelab.io/ipfs/topology/#chart-agent-types-avg
-
Peers without a public IP address continues to be the primary cause of dial errors: https://probelab.io/ipfs/topology/#chart-dial-error-ts
-
DHT Lookup performance is consistently below 1s at the 75th percentile and below 0.5s at the 50th percentile - very reasonable for a P2P network of this size: https://probelab.io/ipfs/dht/#chart-ipfs-dht-lookup-performance-ts
-
Number of “Daily Unique CIDs” has gone down and stabilised to pre-Nov 2025 levels, after a 4x-5x spike during the first half of December 2025: https://probelab.io/ipfs/bitswap/#chart-ipfs-bitswap-unique-cids-ts
-
Blocked requests at the Gateway level showed a small but consistent decline between the 12th and 24th of January, before climbing back up the last couple of days: https://probelab.io/ipfs/gateways/#chart-ipfs-gateway-requests-ts. Still 100% of blocked requests seem to originate from Taiwan.
-
The cid.contact network indexer seems to have recovered from the severe outages earlier in 2025. There is still a non-negligible percentage of delayed ingests at times: https://probelab.io/ipfs/ipni/cid.contact/#chart-ipni-uptime-bar
-
At the same time the number of requests per day has almost doubled since November 2025: https://probelab.io/ipfs/ipni/cid.contact/#chart-ipfs-ipni-requests-ts
-
DHT vs IPNI Lookup comparison: For cached content, IPNI is clearly the fastest (by far). For uncached content, however, the Amino DHT is faster than IPNI (for all regions apart from Asia Pacific). The FullRT version is consistently faster than the Amino DHT, which is expected: https://probelab.io/ipfs/ipni/cid.contact/#chart-ipni-dht-lookup-comparison-cdf
-
DHT vs IPNI Publish comparison: The Amino DHT is by far the slowest here, around 5x slower than the “time to availability” of the cid.contact indexer: https://probelab.io/ipfs/ipni/cid.contact/#chart-ipni-dht-lookup-comparison-cdf
Please respond to this thread, or get in touch with the team directly, in case something looks off, or needs improvement, as well as for general feedback and requests.
The ProbeLab team.