ProbeLab’s Notable IPFS Performance Results - Week 29, 2026

Protocol
1

Following up on our last major update back in Week 18 (late April), we are looking across the data from May and June to bring you the most notable network performance trends up to today (July 13, 2026, Week 29).

This period has seen some fascinating shifts in client behaviour, a massive surge in indexer usage, and a continuing transition towards Service Worker Gateways.

Here’s a summary for the most notable results:

1. Network Topology:

  • DHT Server Baseline: The network’s core routing foundation remains stable. Distinct DHT server IP counts have consistently maintained their baseline, hovering right in the 23k–25k range.

  • The Client Spike Post-Mortem: Back in Week 18, we reported a massive client node spike up to 525k. Over the subsequent weeks, this influx has stabilized, indicating a temporary burst in client nodes that has since plateaued into a steadier network baseline of around 400k client nodes seen over a week.

  • Bootstrappers: Core IPFS bootstrap infrastructure has traditionally been stable and this period has been no different. Zero fluctuations throughout the entire stretch.

  • :bar_chart: Track this here: IPFS Servers vs Clients & Bootstrapper Performance

2. Kubo & DHT Lookup Trends

  • Publish Gains are the New Normal: The phenomenal sub-second publishing times brought on by Kubo v0.39+ have sustained ever since. Content publishing times are holding tight at under 1 second (down from the historic ~14-second averages, prev0.39).

  • Lookup Latency: Average DHT lookup latencies presented no major fluctuation through May and June, maintaining an average just north of 0.3 seconds.

  • :bar_chart: Track this here: Kubo Publish Performance & DHT Lookup Performance

3. Bitswap & Content Volume Growth

  • Sustained CID Volume: The network is continuing to handle a massive volume of unique CIDs. Daily unique CIDs have stayed robustly elevated, consistently pushing over 104M to 120M+ unique CIDs circulated through Bitswap over the past month.

  • The most active peer has made more than 12M requests, while the most popular CID has been requested 343 times, on July 12, 2026 - see last two tables at: Bitswap | ProbeLab Analytics

  • There’s a clear and steady increase in the number of Bitswap WANT CID Requests, with nodes serving close to 80 requests in a 4hr window at the p90.

  • :bar_chart: Track this here: Bitswap Unique CIDs

4. cid.contact Performance Excellence

  • Traffic Influx: The defining trend of 2026 has been the heavy traffic hitting the indexer network and this trend has continued over the last two months. Requests to cid.contact have officially doubled compared to mid 2025, with daily IPNI requests scaling past the 198M mark.

  • Uptime & Ingest Reliability: Despite the massive scale, IPNI uptime has successfully maintained its >99% record. Delayed ingests (advertisements taking longer than 5 minutes to index) have essentially eclipsed.

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

5. Gateways: The Architectural Hand-Off

  • Traditional HTTP Gateway Drop: Public gateway data shows a continuing, clear downward trend in overall traffic hitting traditional gateways like ipfs.io and dweb.link.

  • Service Worker Inroads: This decline correlates directly with the ongoing migration toward Service Worker Gateways.

  • Global TTFB Splits: Looking at Service Worker performance, Time-to-First-Byte (TTFB) for cached content continues a downward trajectory worldwide. However, uncached content (cache misses) has faced structural friction – specifically showing notable TTFB spikes for users routing through Europe and US-East, while remaining relatively steady in alternative global regions. This is an area that could do with some investigation and optimisation.

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

6. Security & Key Distribution

  • Keyspace Density: While the PeerID allocation across the DHT keyspace generally trends safely within bounds, the last few weeks have exposed a few distinct spikes pushing above our designated risk threshold. This indicates localized clustering in certain regions of the keyspace. Although not alarming, they’re worth keeping an eye on.

  • :bar_chart: Track this here: Keyspace Region Population

A shoutout to the infrastructure and maintenance teams for keeping these lines straight while handling escalating indexer traffic. As always, feedback is welcome on this forum or the contact details of the ProbeLab team.

2

Quick note about the Keyspace density

The Keyspace Regions Population chart aggregates at a coarser granularity (depth 6), where each bucket is the sum of 16 finer sub-prefixes at depth 10. So when a depth-6 region crosses the risk threshold, it doesn’t necessarily mean one region is overloaded — it’s more likely that several of its 16 constituent depth-10 prefixes are each modestly above average, and their sum pushes the aggregate over the line.

This is consistent with what we see in the Keyspace Density Distribution at depth=10: no individual 10-bit prefix exceeds ~20 peers. In other words, there’s no single narrow region being disproportionately targeted or unusually dense — the threshold breach at depth 6 reads as a mild, distributed elevation across a handful of sub-prefixes rather than true localized clustering.

Let’s keep an eye on depth 10 for the pattern that would actually be concerning: a single depth-10 prefix starting to dominate its depth-6 region’s population. That’s the signal that would indicate real clustering risk.