Solana streaming explained: how Deshred delivers transactions before execution
Rafael from Triton One works on streaming and Geyser optimisations. He started on Solana in 2021, built a light client called Velos, and got pulled into Triton after a Twitter exchange with Wilfred.
He breaks down how Solana data gets delivered to clients, why the network streams shreds instead of account state, and where the SVM fits in replay.
Solana streaming explained: how Deshred delivers transactions before execution
How Colosseum turned Solana hackathons into a $250K venture pipeline
Realms team on the governance layer on Solana and the Sowellian prediction market
More from Pirates Parley
The web janitor of the Solana Foundation on agents, AEO and building on Solana
Harkle, an ecosystem engineer at the Solana Foundation, joins Pirates Parley to talk about the websites that run Solana, making the web agent-ready, and a years-old argument with Steve about RPC pricing that shaped how Triton charges today. Harkle calls himself the web janitor of the Solana Foundation. He runs solana.com, payments.org and pay.sh, came into crypto through NFTs, and spent years building for the New Zealand government before Solana pulled him in. They cover: Why he calls himself the web janitor of the Solana Foundation Buying his first Bitcoin in 2014 through a Second Life in-game currency Making solana.com agent-ready and ranking for AEO, the agent version of SEO Solana holding around 30% of agent mindshare across the big LLMs The argument with Steve about RPC pricing that flipped his whole thesis Why expensive RPC made him realise Solana was a B2B chain, not P2P Triton's pay-as-you-go model and pay.sh endpoints for agents Two years of 100% Solana uptime since February 2024 Whether the Solana Foundation should work itself out of a job Automating a content pipeline so a Google doc becomes a published page Why he hasn't hand-written a line of code since early 2025 His one rule for getting Foundation support: build a good product Triton One: triton.one/pricing Solana: solana.com/data • pay.sh Harkle: x.com/harkl_
Making Solana fast: from 33% skip rates to 100M compute unit blocks
Alessandro Decina from Anza joins Pirates Parley to break down how the Agave validator client went from constant congestion and 33% skip rates to 100 million compute-unit blocks, 200-millisecond slots, and the Alpenglow consensus rewrite. Alessandro spent 15 years in open-source multimedia and audio-video streaming, wrote one of the early XDP packet-processing implementations in C, and then added eBPF support to the Rust compiler. That work pulled him into Solana, where he now leads performance at Anza on Agave. They cover: Skip rate falling from 33% to 0.5% by version 2.3, and outages becoming a non-event The TPU rewrite to QUIC, stake-weighted QoS and why early spam still got through Why one syscall per packet caps throughput, and how XDP writes straight to the network card The invalidator is an internal team that attacks the cluster to surface the next bottleneck Rewriting parts of the Turbine with XDP so that much larger blocks can propagate Block size moving from 60 million to 100 million compute units, with headroom for more Shorter 200 millisecond slots targeted for the 4.2 release around the end of July Shrinking the leader ingest window from 30 seconds to 2.5 seconds with help from RPC providers Why Alpenglow matters most for deleting years of fragile consensus code A candid take on whether running two clients really improves security The Firedancer rivalry and the offsite that made him commit to making Agave faster His five-year vision of open source, permissionless global finance Triton One: triton.one/pricing Alessandro Decina: x.com/alessandrod Anza: anza.xyz
Bringing privacy to everything on Solana: Arcium
Arcium lets a network run computations on data that stays encrypted from start to finish. The nodes never see what they are processing, not even the operators. Milian, marketing manager at Arcium, joins Steve to explain how that works and what it unlocks. In this episode: The limit of zero knowledge: you can prove you are over 18, but you cannot compute across several parties' data The dishonest majority model, where privacy holds as long as one node stays honest Why TEEs keep getting broken, and where FHE fits next to MPC Training AI on private hospital data without exposing it cSPL, confidential SPL tokens that hide your balance and transfer amounts Why has fully public on-chain wealth become a physical safety problem Try Triton: https://triton.one/pricing Arcium: https://x.com/Arcium Milian: https://x.com/milianstx