You can find all published reports here:

List of Published Reports (Internal)

Overview: Where are we on PeerDAS?

Sunnyside Labs resumed testing the limits of blob throughput under PeerDAS-Devnet-7 conditions.

• Environment: 60 nodes
  • Similar variation but larger than Devnet-7 (36 nodes)
  • Geth/Nethermind/Reth x Grandine/Lighthouse/Lodestar/Nimbus/Prysm/Teku
• Setup: same images on limited resources per node (8GB RAM)
• Approach: Ran two Devnets in parallel, using the Spamoor blob generator to gradually increase blobs per block continuously without resetting the chain

Please find Sunnyside-Devnet-18 & Sunnyside-Devnet-19 dashboards for detailed results.

Key Findings

Network blob throughput repeatedly peaked around 50 blobs per block in test runs. After hitting ~50, performance issues caused slot misses and resets, preventing sustained throughput above this. Multiple spike-and-drop cycles indicate ~50 blobs/block is the current network limit under these conditions.

Compared to previous Devnets

• Previously, only specific client pairs (e.g. Lighthouse/Besu, Lighthouse/Geth) reached 43-50 blobs/block ; others often capped out at 18-35 blobs and saw instability.
• This round, with distributed blob building, reached nearly 50 blobs/block consistently even with a diverse mix of EL/CL combinations.

Latency due to block arrival

Delay from block imports

Delay in attestable slot start time

Bottlenecks: Cumulative delays (block propagation, import & attestation timing) caused slot misses at high load, capping throughput to ~25-35 blobs in some runs

Consensus Stress: At peak blob counts, we saw chain instability - e.g. minor re-orgs and attestation disagreements - signaling the consensus layer under strain.

Supernode Weakness: Supernodes became points of failure at the highest blob loads (details in the next section).