Raw Shreds

Use solana-shred-perf (Solana Shred Latency Benchmark) to compare Raw Shreds and other UDP shred sources on your host.

It uses libpcap to passively capture shred packets — it does not bind the port, so it can run alongside your app, validator, proxy, or relayer without interfering.

For every shred (ShredId = slot + index + type), it records who delivered it first and how far behind everyone else was, then prints periodic latency tables (win rate, p10–p99).

This tool is for Raw Shreds (UDP). For Decoded Shreds, Binary, Preconfs, and Geyser gRPC / Fastlane, use GeyserBench.

What it measures

First source per shred
Delay behind the winner
Win rate per source
p10, p30, p50, p70, p90, p99 delay distribution

Modes

Mode	Use when
`raw-ports`	Multiple deliverers, each on its own destination port
`raw-ips` (recommended for many setups)	Multiple sources send to one shared port; compare by source IP

Build

Requires Rust and libpcap headers.

# Debian/Ubuntu
sudo apt-get install libpcap-dev

git clone https://github.com/shrederxyz/solana-shred-perf.git
cd solana-shred-perf
cargo build --release

Binary: target/release/solana-shred-perf

Run

Packet capture needs elevated privileges (sudo or cap_net_raw).

Compare two ports (`raw-ports`)

sudo RUST_LOG=info ./target/release/solana-shred-perf \
  --interface any \
  --ports 20000,20005 \
  --mode raw-ports \
  --report-interval-secs 20 \
  --settle-window-ms 1000

--interface any captures all interfaces on Linux.

Compare source IPs on one port (`raw-ips`)

sudo RUST_LOG=info ./target/release/solana-shred-perf \
  --interface any \
  --ports 20000 \
  --mode raw-ips \
  --report-interval-secs 20 \
  --settle-window-ms 1000 \
  --label 198.13.137.171=shreder.xyz

A ready-to-edit run.sh is included in the repo.

Useful options

Flag	Default	Description
`--interface`	`any`	Capture interface. `any` captures all interfaces (Linux)
`--ports`	`20000`	Comma-separated UDP destination ports
`--mode`	`raw-ports`	`raw-ports` or `raw-ips`
`--shred-filter`	`all`	`all` or `clean` (`clean` requires `--rpc-url`)
`--rpc-url`	`https://api.mainnet-beta.solana.com`	RPC endpoint for the leader schedule (clean filter only)
`--report-interval-secs`	`10`	Seconds between reports
`--settle-window-ms`	`1000`	Hold shreds before scoring so slower copies can arrive
`--label`	—	`KEY=NAME` friendly row names (repeatable)

Clean shred filter

Some providers send unreconstructable / junk shreds that inflate win rates. With --shred-filter clean, signatures are verified against the slot leader and only verified shreds are scored:

sudo RUST_LOG=info ./target/release/solana-shred-perf \
  --interface any \
  --ports 20000 \
  --mode raw-ips \
  --shred-filter clean \
  --rpc-url https://api.mainnet-beta.solana.com

Reading the table

Matched — finalized shreds this key received
Win % — share of shreds delivered first (higher = faster)
p10–p99 — delay behind the winner (0µs when that key won)
Finalized / Pending — scored vs still inside the settle window

When exactly two keys are present, an A/B summary line is printed.

Benchmarking Shreder

Send Shreder Raw Shreds and your other shred source to the same host.
Use raw-ips (shared port) or raw-ports (separate ports).
Add --label for readable rows.
Compare Win %, p50, p90, and p99 over several report periods.

Repo: github.com/shrederxyz/solana-shred-perf

Methodology article: How to measure raw shred performance

Product: Raw Shreds

On this page