Skip to main content
Once a user has deposited funds onto BULK Exchange on Solana, they are ready to trade. The first step in the high-frequency lifecycle is order ingress, which is handled by the custom-built BULK Net propagation layer. This layer is engineered to solve one of the most fundamental challenges in distributed systems: disseminating information to all participants as quickly and reliably as possible. BULK Net employs a powerful technique from information theory called Reed-Solomon erasure coding to ensure data integrity and resilience against network packet loss. This can be understood with a simple analogy. Imagine you need to communicate the precise shape of a parabola to a group of friends. A unique parabola can be defined by just three distinct points. If you send only those three points, the loss of even a single point during transmission would make it impossible for your friends to reconstruct the original curve. Instead, you can use the equation of the parabola to calculate two additional points that also lie on the curve, and send all five points. Now, even if any two of the five points are lost, your friends can simply pick any three of the points they did receive and perfectly reconstruct the exact same original parabola. BULK Net applies this same principle to order data. A batch of orders, typically around 256 orders totaling approximately 20 KB, is treated as the “parabola.” This 20 KB data block is mathematically encoded into 8 “shards,” consisting of 6 data shards and 2 parity shards (a configuration denoted as   k=6+2). Each shard is a tiny packet of about 3 KB. The magic of this encoding is that any validator that receives any 6 of the 8 shards can perfectly and unambiguously reconstruct the original 20 KB order batch. This provides immense fault tolerance; the network can sustain significant packet loss without losing any order data. UDP Fan-Out: Optimized for Latency Once the order batch is encoded into shards, it must be broadcast to the entire validator network. BULK Net achieves this with a two-hop fan-out mechanism that leverages raw UDP for maximum speed. The process is as follows:
  • First Hop: The validator that initially receives the order (Validator A) immediately transmits the 8 tiny shards via raw UDP to 8 of its peers, selected based on their stake weight.
  • Second Hop (Immediate Relay): Crucially, each of these 8 peers does not wait to receive all the shards. The moment a peer receives its first shard from Validator A, it immediately begins forwarding that shard to its own set of 8 stake-weighted peers.
This design, inspired by the Rotor data propagation protocol in Solana’s Alpenglow, is meticulously optimized for low latency. The performance gains come from several key factors:
  • Tiny Payloads: Transmitting 3 KB shards is far more efficient than broadcasting large, 192 KB blobs as seen in Solana’s Turbine protocol.
  • Elimination of Forwarding Delay: The “immediate relay” mechanism prevents the “receive-full-blob-then-verify-to-forward” delay common in other gossip protocols. Latency is not cumulative at each hop.
  • High Stake bandwidth: Majoritiy of solana stake is concentrated in key bandiwth regions that makes propagation known (not blind) and fast