1 of 10

Slot Timings

Preliminary Results

Maria Silva

ACDC #161

July 24, 2025

Researcher @ EF

ACDC #161

Slot timings - prelim results

Maria Silva - EF

2 of 10

What are we measuring?

0s

12s

4s

8s

Propagation

Execution + validation

Attestation propagation

Attestation aggregation

Aggregate propagation

Block arrivals minus relay publishing time

Attestation arrival minus relay publishing time

Attestation arrival after 4s for missed slots

Methodology

ACDC #161

Slot timings - prelim results

Maria Silva - EF

3 of 10

Data Sources

Methodology

  • Block and attestation events from libp2p:
    • Prysm fork from ethpandaops team (in Xatu)
    • San Francisco, Amsterdam, Bangalore & Sydney
    • 3 nodes per location subscribed to committees 0 and 1
  • Block events from community nodes:
    • Collected from Xatu’s data (contributed by the community)
    • Varied locations and clients
  • Block publish times by Ultrasound relay
    • Thanks for the team for sharing the data!

ACDC #161

Slot timings - prelim results

Maria Silva - EF

4 of 10

Block propagation - libp2p table

Results

95th Percentiles:

Australia - 993 ms

India - 832 ms

Netherlands - 529 ms

United States - 706 ms

High bandwidth + block distributed by relay

ACDC #161

Slot timings - prelim results

Maria Silva - EF

5 of 10

Block propagation - beacon table

Results

More prone to errors as it is open to the community

95th Percentile: 2979 ms

ACDC #161

Slot timings - prelim results

Maria Silva - EF

6 of 10

Attestations - arrivals after relay

Results

95th Percentiles:

P65 - 3397 ms

P95 - 4413 ms

High bandwidth + block distributed by relay

ACDC #161

Slot timings - prelim results

Maria Silva - EF

7 of 10

Pure attestation propagation - missed slots

Results

Small sample + attestation concentration delays

ACDC #161

Slot timings - prelim results

Maria Silva - EF

8 of 10

Overview

0s

12s

4s

8s

Propagation

Execution + validation

Attestation propagation

Attestation aggregation

Aggregates propagation

95th percentiles:

993 ms from libp2p worse location

2979 ms from community nodes

P95 within committee has a 4413 ms 95th percentile

Average P90 is 1641ms → need to investigate what is happening in the remaining 10%

Results

ACDC #161

Slot timings - prelim results

Maria Silva - EF

9 of 10

Next steps

Next steps

  • Collect more data - relays, community data for attestations, bigger samples
  • Additional analysis
    • Blobs propagation
    • Aggregations
    • Investigate arrivals for locally-built blocks
    • Investigate arrivals for small blocks (in size or gas)
      • How does block size affect block propagation?
      • How does gas used affect execution?
    • Investigate late arrivals - what factors contribute to late arrivals?

ACDC #161

Slot timings - prelim results

Maria Silva - EF

10 of 10

Thank you!!

Question? Comments?

ACDC #161

Slot timings - prelim results

Maria Silva - EF