Stereum (Home of StereumLabs) @stereumdev

Powering Ethereum infrastructure. Home of StereumLabs — AI-powered observability for Ethereum nodes. Plus Stereum OSS · StereumPlus · Fungl. stereumlabs.com Austria Joined April 2014

Tweets

696
Followers

2K
Following

95
Likes

626

Lido @LidoFinance

a month ago

Join the Lido Community Call next Thursday. Catch up on what’s next for the Lido staking modules, Lido's recent Web3SOC certification, and guest sessions on client diversity. Add it to your calendar: luma.com/pmb473ba

10 10 51 10K 2

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Full breakdown: per-EC pruning observability, the Nethermind v1.37.1 inline-newPayload trade-off, Ethrex with zero pruning metrics, Reth still in initial staged sync, EIP-4444 implications. docs.stereumlabs.com/blog/ethereum-…

0 0 0 39 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Sync-committee duties get hit before attestations. A late sync-committee message misses the next block's sync_aggregate outright. An attestation still has one slot of grace. Today at @ethStaker's Fast Finality call (15:00 UTC), three of five ECs on Nimbus cross the deadline.

1 0 0 50 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Five synced ECs on the same Ethereum mainnet: Erigon: ~509 GB Ethrex: ~598 GB Nethermind: ~1,225 GB Besu: ~1,261 GB Geth: ~1,688 GB 3.3× spread, driven by retention architecture. Numbers are upper bounds (hosts rebuilt in 7-16 days). And disk is only half of it.

1 3 4 148 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

@fede_intern We measure exactly that misunderstanding gap. Per-node reality on our fleet: a Besu node's 24 peers are 75% Reth. A 12% slice of the network becomes 75% of one node's reality. Aggregate diversity numbers hide this entirely. docs.stereumlabs.com/blog/ec-p2p-pe…

0 0 0 4 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

@ethStaker @benjaminion_xyz @robsaltini @Butta_eth Adjacent data point for the call: same EC behind two CCs produces 2-5× different reorg counts on identical hardware. Re-baseline alert thresholds per CC × EC pair. Lodestar reports 0 reorgs across 90 days while Prysm increments 1,002× per host. docs.stereumlabs.com/blog/ethereum-…

0 0 0 38 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Full breakdown: 6 ECs, 130-to-24 peer-count spread, 50,000× churn-rate spread, eth68 hard caps on Erigon, and why DevP2P has no diversity-aware peer scoring like gossipsub does. docs.stereumlabs.com/blog/ec-p2p-pe…

0 0 0 26 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Adjacent finding: same Geth on bare-metal (port 30303 open) = 50 peers. On GCP (default firewall, port blocked) = 15 peers. CC peer count looks fine throughout. The EC is silently running 70% degraded. Lots of cloud nodes probably are.

1 0 0 60 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Aggregate Ethereum client diversity: 'no client over 55%.' The view from an average Besu node on our fleet: 75% of its 24 peers are Reth. A 12% slice of the network becomes 75% of one node's reality.

1 4 5 147 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Stereum Launcher v2.4.8 is out. - Ubuntu 26.04 support Auto-updates on startup. MacOS users: download the dmg manually and enable via System Settings. github.com/stereum-dev/et…

0 4 6 104 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Three more findings in the 90-day census: - Geth is the only EC whose Prometheus reorg counter ever increments - Nethermind & Reth export it but it never moves - Besu, Erigon, Ethrex don't export one at all docs.stereumlabs.com/blog/ethereum-…

0 0 0 56 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

It gets worse in the matrix. The same EC behind two CCs produces 2-5× different reorg counts: - Prysm + Besu: 69 per host - Prysm + Nethermind: 280 per host A fixed alert threshold doesn't port between pairings. Re-baseline after every EC change.

1 0 0 68 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Same chain. Same bare-metal. Same metric name: beacon_reorgs_total. Prysm increments it 8× more often than Lighthouse over 7 days. Lodestar: 0 increments across 90 days. Both readings are correct. The clients just count different things.

1 4 6 132 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Quiet Saturday on the fleet. 90 hosts across Vienna and GCP humming along. ~30,000 log lines per second flowing into Elasticsearch. The boring part of this job is what makes the interesting part possible.

0 1 3 46 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Client diversity is a CROPS precondition. We measure all 36 EC × CC pairings on identical bare-metal. Same-protocol spreads of 10 to 50× are routine. We publish the numbers so 'diversity' stays measurable.

vitalik.eth @VitalikButerin

2 months ago

Some of my perspective on where the @ethereumfndn is going. First of all, this is only my own view. The board is not just me, and I have no extra special powers on the board that the other board members do not. @aerugoettinea is the one executing much of this transition. My

2K 2K 8K 4.2M 3K

0 3 3 120 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

There's more in the full post: a 43% data column success rate on Prysm, a 6.5× spread between Reth and Erigon on blob serving, and a Teku custody-group config gap worth checking. docs.stereumlabs.com/blog/blob-perf…

0 0 2 77 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

The reason: Lodestar is the only TypeScript-based consensus client. KZG and hash ops are optimized in compiled languages: Rust (Lighthouse, Grandine), Go (Prysm), Java/JNI (Teku), Nim (Nimbus). A JIT runtime hits a hard ceiling here.

1 0 2 133 0

View Details

Stereum (Home of StereumLabs) @stereumdev

2 months ago

Blob verification on Ethereum varies 50× between consensus clients. Teku does it in 3.9ms (p50). Lodestar takes 469ms (p95). Same bare-metal, same network. At Teku's pace the attestation deadline is no problem. At Lodestar's, it gets tight.