ProbeLab’s Notable IPFS Performance Results - Week 36, 2026

Protocol
, , ,
1

Following up on our update back in Week 29 (mid-July), we are looking across the data from July and August to bring you the most notable network performance trends up to today (Week 36, September 2026).

This period saw a significant mid-summer influx of DHT client nodes, sustained sub-second publish times, continuous high volume on indexers, but also stabilization of some core network metrics and smooth gateway migration.

Here is a summary of the most notable results:

1. Network Topology

  • DHT Server Baseline: Core DHT server nodes remain remarkably stable, continuing to hover steadily around the 23k–25k mark in terms of DHT Server IP addresses. The actual DHT Server nodes, in terms of PeerIDs seen is also steady, around 12k-13k. Both of these numbers refer to the weekly average over all crawls. The number of servers at a specific point in time (network snapshot) stands at 4k nodes simultaneously online.
  • Mid-August Client Spike: In contrast to the stable server count, the estimated number of active Amino DHT client nodes experienced a significant surge in mid-August (Weeks 33 and 34), peaking at ~800k client nodes (up from the post-Week 18 baseline of ~400k). Over late August and early September, this spike has begun to taper back down toward normal levels.
  • Bootstrappers: The IPFS Amino DHT bootstrap infrastructure maintained 100% availability with zero performance degradation or churn during the client traffic influx.
  • :bar_chart: Track this here: IPFS Servers vs Clients & Bootstrapper Performance

2. Kubo & DHT Lookup Trends

  • Publish Gains holding strong: Content publishing times introduced with Kubo v0.39+ have sustained their sub-second performance, holding tight at under 1 second across the network.

  • Lookup Latencies: Despite the temporary doubling of active client nodes in mid-August, DHT lookup latencies remained unaffected, maintaining a steady average around 0.3s–0.35s.

  • Regional Lookup Performance (IPNI vs. DHT): Comparing lookup performance across geographic probe locations reveals a distinct latency gradient:

    • Europe (eu-central-1) consistently records the fastest lookup times across both IPNI and Amino DHT resolutions.

    • Latency grows progressively slower as requests move across to US regions (us-east-2, us-west-1), and peaks in APAC (ap-southeast-2).

    • This trend highlights the higher concentration of responsive indexer nodes and DHT server infrastructure closer to European regional hubs, resulting in noticeably tighter latency CDF curves for European clients compared to APAC.

  • :bar_chart: Track this here: Kubo Publish Performance, DHT Lookup Performance, & IPNI vs DHT Lookup Comparison CDF

3. Bitswap & Content Volume Growth

  • Sustained CID Volume: Daily unique CIDs circulated through Bitswap stayed strongly elevated throughout August, consistently pushing over 100M to 120M+ unique CIDs/day.

  • WANT CID Demand: The p90 WANT CID request volume closely mirrored the mid-August node influx, showing increased peer interactions during the client activity peak.

  • :bar_chart: Track this here: Bitswap Unique CIDs

4. cid.contact Performance Excellence

  • Indexer Ingress: Traffic hitting cid.contact remained near historical highs, handling over 198M+ daily IPNI lookup requests, especially during August.

  • Ingest Uptime: The IPNI infrastructure maintained its >99% uptime threshold throughout the period, with advertisement ingestion delays (>5 min), or downtime remaining practically non-existent. Impressive performance!

  • :bar_chart: Track this here: IPNI Uptime & Latencies & IPNI Requests

5. Gateways & Service Worker Performance

  • Massive Traffic Spike: Public HTTP gateway requests (ipfs.io, dweb.link) experienced a massive volume surge alongside the mid-August client node influx, peaking at several billion requests per day (a multi-fold increase over the baseline 200M–300M daily average seen in prior months).

  • Correlated Activity & Rate Limiting: This multi-billion daily request surge indicates heavy automated/bot polling or specialized client scraping activity hitting gateway endpoints in tandem with the DHT client node increase. A notable portion of this volume was handled via CloudFlare security filtering and HTTP 429 rate-limiting actions before scaling back down toward baseline levels in late August.

  • Service Worker Adoption (inbrowser.link): Despite temporary traffic surges on legacy public gateways, the structural migration toward client-side verification via Service Worker Gateways continues to progress steadily.

  • Service Worker TTFB Performance:

    • Cached Content (Warm State): Global median TTFB for cached content remains low, hovering tightly between ~150ms–250ms.

    • Uncached Content (Cache Misses): Regional measurements reveal noticeable latency tails:

      • US-West & Asia Pacific: Stable performance averaging ~0.8s–1.2s.

      • Europe & US-East: Exhibited elevated TTFB on cache misses, with median times between 1.8s–2.5s (and p90 tails reaching 3.5s+). ProbeLab is actively tracking connection establishment overhead in these regions.

  • :bar_chart: Track this here: IPFS Gateway Requests Time-Series & Service Worker Gateway TTFB

6. Security & Keyspace Distribution

  • Keyspace Density & Prefix Depth: As noted in the discussion following Week 29, region threshold crossings at depth-6 reflect mild distributed elevations across sub-prefixes rather than concentrated clustering. At depth-10, no single prefix exceeded ~20 peers, indicating no active targeted key-clustering attack or localized routing table overload.

  • :bar_chart: Track this here: Keyspace Region Population

Huge thanks to the network infrastructure and developer teams maintaining performance across all fronts! Feel free to share your thoughts, questions, or observations below, or reach out to the team via the ProbeLab contact page.

1 Like