latency-monitor.sh

#!/usr/bin/env bash
# Bash-only Celestia blob submit load tester.
# Usage: COUNT=100 NS=0x676d DATA=0x676d CMD="celestia blob submit" ./celestia_blob_bench.sh

set -euo pipefail

COUNT=${COUNT:-20}
CMD=${CMD:-"celestia blob submit"}
NS=${NS:-"0x676d"}
DATA=${DATA:-"0x676d"}

printf "Running %s submissions: %s %s %s\n" "$COUNT" "$CMD" "$NS" "$DATA"

ok=0
fail=0
times=()

has_jq=1
if ! command -v jq >/dev/null 2>&1; then
  has_jq=0
  echo "Note: jq not found; height/commitment parsing skipped." >&2
fi

for i in $(seq 1 "$COUNT"); do
  start_ns=$(date +%s%N)
  out=$($CMD "$NS" "$DATA" 2>&1) || rc=$?
  end_ns=$(date +%s%N)
  dur_ms=$(( (end_ns - start_ns) / 1000000 ))

  rc=${rc:-0}
  if [ "$rc" -eq 0 ]; then
    ok=$((ok + 1))
    times+=("$dur_ms")
    if [ "$has_jq" -eq 1 ]; then
      height=$(echo "$out" | jq -r '(.result.height // .height // "-")')
      commitment=$(echo "$out" | jq -r '(.result.commitments[0] // .commitments[0] // "-")')
    else
      height="-"
      commitment="-"
    fi
    printf "[%03d] ok   %7.2f ms | height=%s | commitment=%s\n" "$i" "$dur_ms" "$height" "$commitment"
  else
    fail=$((fail + 1))
    printf "[%03d] fail %7.2f ms | rc=%s | %s\n" "$i" "$dur_ms" "$rc" "$out"
  fi
  unset rc
done

if [ "${#times[@]}" -gt 0 ]; then
  sum=0
  min=${times[0]}
  max=${times[0]}
  for t in "${times[@]}"; do
    sum=$((sum + t))
    (( t < min )) && min=$t
    (( t > max )) && max=$t
  done
  avg=$(printf "%s\n" "${times[@]}" | awk '{s+=$1; c++} END {if(c>0) printf "%.2f", s/c; else print "0"}')
  printf "\nSummary: total=%s ok=%s fail=%s\n" "$COUNT" "$ok" "$fail"
  printf "Durations (ms): avg=%s min=%s max=%s\n" "$avg" "$min" "$max"
else
  printf "\nSummary: total=%s ok=%s fail=%s (no successes to compute stats)\n" "$COUNT" "$ok" "$fail"
fi

Celestia Blob Submit Benchmark Comparison

Runs: 20 submissions each
Command: celestia blob submit 0x676d 0x676d

Average Latency

• Run 1 (rpc-mocha.pops.one): 12,597.20 ms
• Run 2 (full.consensus.mocha-4.celestia-mocha.com): 15,088.35 ms
• Run 2 is ~19.78% slower on average

Min / Max Latencies

• Run 1: min 11,520 ms · max 19,713 ms

• Run 2: min 10,654 ms · max 58,556 ms

  • Run 2 shows a major outlier (58.5s), causing higher variance.

Summary

• Typical per-submission latency (11–13s) is similar in both runs.

• The primary difference is consistency:

  • Run 1 is stable.
  • Run 2 suffers from a large tail latency event, increasing the average by ~20%.

📊 Updated Stats (Removing the 58,556 ms Outlier)

Averages

• Run 1 avg (rpc-mocha.pops.one): 12,597.20 ms
• Run 2 avg (full.consensus.mocha-4.celestia-mocha.com) (outlier removed): 12,800.58 ms

Difference

• Absolute: +203.38 ms
• Percentage: ~1.61% slower

➡️ Once the outlier is removed, the two runs perform nearly identically, with Run 2 only ~1.6% slower.

---

📈 Graph (Run 1 vs Run 2 Without Outlier)

The graph below was generated from your data and shows latencies per submission:

from celestia-app usinggo run ./tools/latency-monitor -a [YOUR_ACCOUNT_MONIKER] -b 1048000 -z 1048000 -d 1500ms -e "grpc-mocha.pops.one:9090"

where my account is: https://mocha.celenium.io/address/celestia1mh0hxmde0mcpx4zhzlt0wua4490fhzjzsc2w6r?tab=transactions

Transaction Statistics:
Total transactions: 294
Successful: 294 (100.0%)
Failed: 0 (0.0%)

Latency Statistics (successful transactions only):
Average latency: 12406.89 ms
Standard deviation: 3826.13 ms

using go run ./tools/latency-monitor -a [YOUR_ACCOUNT_MONIKER] -b 1048000 -z 1048000 -d 1500ms -e "full.consensus.mocha-4.celestia-mocha.com:9090"

with the same account:

Transaction Statistics:
Total transactions: 293
Successful: 293 (100.0%)
Failed: 0 (0.0%)

Latency Statistics (successful transactions only):
Average latency: 12133.84 ms
Standard deviation: 4082.07 ms

Latency Summary (grpc-mocha.pops.one vs full.consensus.mocha-4)

• grpc-mocha.pops.one

  • Avg latency: 12,406.89 ms
  • Std dev: 3,826.13 ms

• full.consensus.mocha-4

  • Avg latency: 12,133.84 ms
  • Std dev: 4,082.07 ms

Comparison

• full.consensus.mocha-4 is ~273 ms faster on average
• This equals about 2.2% lower latency
• However, it also has slightly higher jitter (larger std deviation)

Bottom Line

Both endpoints perform similarly, but full.consensus.mocha-4 is marginally faster while grpc-mocha.pops.one is slightly more stable.