The fastest thing we can send you is the thing we received
Solana broadcasts shreds as UDP datagrams. Anything that hands you something else has decoded, re-framed and re-transmitted them first, and every one of those steps costs time. We forward the original packets because there is nothing faster to forward.
What UDP actually gives you
No handshake, so a packet can be sent the instant it exists. No ordering guarantee, so one delayed packet does not hold up the packets behind it. No retransmission, so nothing is ever re-sent at the cost of delaying what is current.
That last property is the one people find counter-intuitive. TCP's head-of-line blocking means a single lost segment stalls the entire stream until it is recovered, and for market data an old packet arriving on time is worth less than a new packet arriving now. Solana already builds recovery into the data itself with erasure coding, so the transport does not need to solve a problem the protocol has already solved.
What to build for
Every fast market data feed in the world works this way, and the engineering around it is well understood. Here is what a good receiver accounts for, and our documentation covers each one in depth.
Size the receive buffer
Solana builds erasure recovery into the data itself, so a well sized buffer keeps you whole through the bursts. Our calculator does the arithmetic for you.
Watch the counters
The kernel tracks exactly what your socket sees. Poll it, alert on it, and you always know your feed is healthy.
Keep the reader hot
A tight read loop that hands off immediately is the whole trick, and it is a couple of hundred lines. Our reference implementation is annotated end to end.
Decode on your terms
You own the parsing, which means you can strip out everything you do not care about and go faster than any generic feed would let you.
Who this is built for
Teams who want the earliest observable form of on-chain activity and intend to do something specific with it. If any of these sound like you, this is the feed you want:
You are running or building latency sensitive strategies and every millisecond in the path is worth money.
You already have a decoder, or you want one you control rather than one somebody else's schema decides for you.
You want the full stream rather than a filtered subscription somebody else defines.
You would rather pay one flat price than a bandwidth bill that grows with your success.
You want your data arriving on your own hardware, in your own network, under your own rules.
Curious how we stack up on price? The comparison pages put us next to everyone else in the market.
Why this wins on latency
Every layer between the validator and your process is a tax you pay on every single packet, forever. The 1228 byte datagram we forward is the earliest observable form of what is happening on chain, and we hand it to you with nothing in between.
That is the entire product, and it is why one flat $100 buys the same stream that costs multiples of it elsewhere.