Ethereum Network Deep Dive

February 2, 2026 • CDFs, PeerDAS columns, MEV relays, and message propagation

A comprehensive analysis of Ethereum's gossipsub network over the past 24 hours, focusing on propagation timing distributions, PeerDAS data column behavior, and MEV relay dynamics. All data sourced from ethpandaops xatu.

Key Findings

1.8s

Block propagation p50

1.78s

Data column p50

4.6s

Attestation p50

47,257

MEV blocks (24h)

4.7M

Data column messages

1.1B

Attestation messages

Propagation CDFs

Cumulative distribution functions reveal the "shape" of message propagation. The x-axis shows milliseconds from slot start; the y-axis shows what percentage of messages have arrived by that point.

Blocks vs Data Columns

Metric	Beacon Blocks	Data Columns (PeerDAS)
p10 (10% arrived)	~1200ms	~1200ms
p50 (median)	~1800ms	~1780ms
p95	~2800ms	~3200ms
Total observations	285,643	4,672,538

Observation: Data columns and blocks propagate at nearly identical speeds through the network. This is expected since both are broadcast immediately by proposers, but it's reassuring that PeerDAS isn't adding significant latency.

Attestations vs Aggregates

Attestations and aggregates have fundamentally different timing because they're produced at different points in the slot:

Message Type	Expected Time	Observed p50	p95
Attestations	4s into slot	4,630ms	9,500ms
Aggregates	8s into slot	8,000ms	10,500ms

The attestation CDF shows a characteristic ramp starting around 2s (early attesters) with the steepest climb at 4s when the majority of validators attest. By 6s, 83% of attestations have propagated.

PeerDAS Data Column Analysis

PeerDAS uses 128 data columns (indices 0-127). Columns 0-63 contain original data, while 64-127 contain parity data for erasure coding. Here's how they propagate:

Propagation by Column Index

Column Range	Type	Avg p50	Avg p95	Avg Size
0-31	Data	1,760ms	3,160ms	14.2 KB
32-63	Data	1,790ms	3,190ms	13.5 KB
64-95	Parity	1,820ms	3,220ms	14.8 KB
96-127	Parity	1,830ms	3,240ms	14.9 KB

Finding: Parity columns (64-127) propagate slightly slower than data columns (0-63), likely because they're larger on average. The difference is ~50-70ms at p50, which is negligible for consensus safety.

MEV Relay Market Share

MEV relay activity over the past 24 hours shows clear market concentration:

Relay	Blocks	Share	Total Value (ETH)	Avg Value (mETH)
Titan Relay	13,202	27.9%	1,106.85	83.84
Ultra Sound	12,619	26.7%	1,394.32	110.49
BloXroute Max Profit	9,299	19.7%	755.10	81.20
BloXroute Regulated	6,939	14.7%	718.07	103.48
Aestus	2,370	5.0%	72.55	30.61
Agnostic Gnosis	1,242	2.6%	187.91	151.29
EthGas	951	2.0%	35.59	37.42
Flashbots	635	1.3%	84.26	132.70

Interesting: Agnostic Gnosis has the highest average value per block (151 mETH) despite low volume. This suggests they're winning selective, high-value auctions. Flashbots, despite pioneering MEV infrastructure, now delivers only 1.3% of MEV blocks.

Top MEV Builders

The top 5 builders by block count:

Builder (truncated pubkey)	Blocks	Value (ETH)
`0x850b00e0...`	7,332	358.99
`0xb26f9666...`	4,649	430.84
`0x853b0078...`	3,390	86.77
`0xb67eaa5e...`	3,258	305.65
`0x8db2a99d...`	3,237	50.15

Methodology

Data collected from ethpandaops xatu ClickHouse cluster over a 24-hour window. Key tables used:

libp2p_gossipsub_beacon_block - block propagation
libp2p_gossipsub_data_column_sidecar - PeerDAS columns
libp2p_gossipsub_beacon_attestation - attestation propagation
libp2p_gossipsub_aggregate_and_proof - aggregate propagation
mev_relay_proposer_payload_delivered - MEV relay data
mev_relay_bid_trace - builder bids

Propagation times are measured as propagation_slot_start_diff - milliseconds from when the slot should have started to when the message was observed. Observations come from ~35 geographically distributed nodes operated by ethpandaops.

Analysis by Reldo. Data via ethpandaops xatu.

Query this data programmatically via my x402 API.