You can find all published reports here:

List of Published Reports

Overview

This week's testing concentrated on identifying bottlenecks and potential improvements, specifically aimed at achieving higher blobs per block. Initial tests removed Lodestar and Nimbus to determine their impact on blob throughput. Subsequently, consensus-layer (CL) specific tests were conducted using Geth as the execution layer (EL), particularly assessing the performance impacts of enabling or disabling GetBlobsV2 across different CL clients.

Performance

1. Combination of ELs/CLs

Devnet-20 Blobs/Block

Devnet-20 Blobs/Block

Devnet-20 Block Time

Devnet-20 Block Time

A test was conducted with a mix of EL and CL pairs (Grandine, Lighthouse, Prysm, Teku combined with Geth, Nethermind, and Reth). Each client pairing included five nodes, totaling 60 nodes. The test achieved a maximum of approximately 40 blobs per block, notably lower than previous tests involving Lodestar and Nimbus. Despite lower peak performance, the overall network stability improved, including better block propagation times.

2. Lighthouse-Geth

Blobs/Block on Lighthouse-Geth GetBlobsV2 On

Blobs/Block on Lighthouse-Geth GetBlobsV2 On

Block Time on Lighthouse-Geth GetBlobsV2 On

Block Time on Lighthouse-Geth GetBlobsV2 On

Blobs/Block on Lighthouse-Geth GetBlobsV2 Off

Blobs/Block on Lighthouse-Geth GetBlobsV2 Off

Block Time on Lighthouse-Geth GetBlobsV2 Off

Block Time on Lighthouse-Geth GetBlobsV2 Off

A dedicated test on 50 Lighthouse-Geth nodes was performed using Spamoor blob stress tests (three rounds in a row). Tests revealed that enabling GetBlobsV2 significantly improved performance, allowing around 53 blobs per block compared to approximately 45 without the feature enabled. Despite higher performance, the tests with GetBlobsV2 enabled experienced increased verification failures.

3. Prysm-Geth

Blobs/Block on Prysm-Geth GetBlobsV2 On

Blobs/Block on Prysm-Geth GetBlobsV2 On

Block Time on Prysm-Geth GetBlobsV2 On

Block Time on Prysm-Geth GetBlobsV2 On