You can find all published reports here:
Sunnyside Devnet Reports (Internal)
In Collaboration with EthPandaOps 🐼 💙
Nearly 1700 nodes were deployed to simulate ~1/8th of Mainnet this time, Sunnyside & Base providing almost 800 nodes. We believe running at such scale would give one of the closest approximation to blob scaling before Fusaka.
We aim to provide our analysis and perspectives on blob scaling: blob throughput, bottlenecks and HW requirements in addition to PandaOps’ in-depth analysis. PandaOps’ analysis gives strong basis of reasons why we can do early BPOs with high level of confidence, but they found attestation becoming less stable from 20 blobs or higher leaving doubts to scale more than 20 blobs in later BPOs.
As many have anticipated, we see meeting network requirements from EIP-7870 as one of the main bottlenecks alongside others, making difficult to attest in time. Meanwhile, EF P2P team has been suggesting that we could be able to achieve higher blob counts with more stability with few non-fork related changes.
Let’s explore into such bottlenecks and new findings from testing at scale compared to smaller devnets to help shaping further BPO schedule.
Below is the setup of Sunnyside Lab’s nodes. Please find https://notes.ethereum.org/@ethpandaops/fusaka-devnet-5 for overall spec.
Validating Nodes | Non-Validating Nodes | |
---|---|---|
Node combination | All 390 Lighthouse & | |
196 Teku DigitalOcean Fullnodes | All 180 Non-Validating Nodes | |
(5 x 6 CLs x 6 ELs) | ||
Hardware | 8 vCPUs / 16GB RAM / NVMe SSDs | 4 vCPUs / 8GB RAM / NVMe SSDs |
Bandwidth Cap | 50/25Mbps (PandaOps had 100/50) | 50/15Mbps |
Validator distribution | 8 validators per node | N/A |
Supernode distribution | None from Sunnyside Labs | N/A |
Images: ‣