22 · senior eng @triton_one · fastest streaming on @solana · low-latency systems & networking · @superteamBR · ibrl

Joined September 2021
406 Photos and videos
Leafar 🦀 retweeted
faster slot switches when leader incredibly faster accountsdb faster epoch switch 5x faster p99 tx execution 4.2 is going to be another monster release 🤯
Replying to @anza_xyz
2/ some recent-ish wins in master: ptoken scheduler bindings restart speedup async replay verify direct mapping 200ms slots TxV1 10x rent reduction XDP & 100M CUs better repair Alpenglow
3
7
53
2,423
Leafar 🦀 retweeted
It launches with two open-source projects from @triton_one: Superbank — Solana's entire 400B transaction ledger, queryable in milliseconds. Up to 38x faster historical reads. blog.triton.one/introducing-…
1
2
42
1,740
Leafar 🦀 retweeted
Introducing the Solana RPC Working Group Its mission is to advance Solana's open-source RPC stack, making the read layer more accessible for all. Built in the open by @triton_one, @Helius, @JupiterExchange, @anza_xyz, and @SolanaFndn 🧵
15
43
263
24,093
Leafar 🦀 retweeted
Replying to @anza_xyz
2/ some recent-ish wins in master: ptoken scheduler bindings restart speedup async replay verify direct mapping 200ms slots TxV1 10x rent reduction XDP & 100M CUs better repair Alpenglow
4
12
66
4,397
Leafar 🦀 retweeted
Solana's read layer 2.0 is now open source: Superbank (history) and Cloudbreak (accounts), marking the end of vendor lock-in for chain access. Querying @solana should feel like calling a real database, so we built exactly that. 🟣Lightning-fast account reads with dynamic indexes from your query patterns 🟣Your own queryable copy of Solana history, with no indexing logic to write 🟣More performant than Agave, while running on commodity hardware Fork it, run it, improve it. It's yours 🌊🧵x.com/solana_devs/status/206…
Introducing the Solana RPC Working Group Its mission is to advance Solana's open-source RPC stack, making the read layer more accessible for all. Built in the open by @triton_one, @Helius, @JupiterExchange, @anza_xyz, and @SolanaFndn 🧵
13
27
107
9,085
Leafar 🦀 retweeted
I won’t rest umtil we make @SuperteamBR the #1 CNN Brasil went to Solana house and interviewed our legends. Was not easy to get those guys 🔥🔥🔥
5
3
35
811
Leafar 🦀 retweeted
Hey, we’ve open sourced our ledger stack at @triton_one! It’s called Superbank, and it’s pretty awesome. 100% open-source under AGPLv3. Give it a try and let me know what you think! github.com/solana-rpc/superb…
11
13
71
3,131
Leafar 🦀 retweeted
When the command line flags just aren't cutting it for your network setup; do the work, submit the patches, get it in prod. XDP live on validators @triton_one manages now.
100M CU blocks need @solana validators relaying shreds faster than they do today. XDP is how they get there: it speeds up Turbine's outbound propagation 750x, ensuring every block reaches the whole cluster before the next slot starts. We moved our entire fleet to XDP, and to help other operators get there, we upstreamed two patches to @anza_xyz's Agave, removing common adoption blockers: → VLAN tagging in userspace for segmented datacenter networks. Operators who split traffic across VLANs can now turn on XDP without the switch dropping their shreds → Multicast MAC derivation unlocks multicast retransmit on XDP for group sends. Useful on @DoubleZero and anywhere per-peer duplication isn't viable If there's a validator you'd like to see on XDP, tag them below👇
3
7
257
Leafar 🦀 retweeted
Our way is to not just write technical content for our blog bragging about how fast we are. Rather we write code, make it open source, improve things for everyone and *then* write technical content for our blog.
Replying to @solana
1. github.com/anza-xyz/agave/pu…: VLAN tagging in userspace 2. github.com/anza-xyz/agave/pu…: multicast MAC derivation per RFC 1112 By @leafar_dev 💜
4
16
953
Leafar 🦀 retweeted
What does it take to train an AI model without ever seeing the data it learns from? That's one of the questions @SteveCleanBrook will ask @cryptopapi997 (CTO and co-founder of @Arcium) on this week's Pirates Parley about privacy, encryption, and multi-party computation. And because great questions make the best shows, we're rewarding the top three with $50 in RPC credit. To participate, quote RT this post with yours and ask it live on stream: 🗓️Wednesday 10/06 at 3 PM UTC 📍Live on X and YouTube
6
8
41
12,761
xdp is good 👍
100M CU blocks need @solana validators relaying shreds faster than they do today. XDP is how they get there: it speeds up Turbine's outbound propagation 750x, ensuring every block reaches the whole cluster before the next slot starts. We moved our entire fleet to XDP, and to help other operators get there, we upstreamed two patches to @anza_xyz's Agave, removing common adoption blockers: → VLAN tagging in userspace for segmented datacenter networks. Operators who split traffic across VLANs can now turn on XDP without the switch dropping their shreds → Multicast MAC derivation unlocks multicast retransmit on XDP for group sends. Useful on @DoubleZero and anywhere per-peer duplication isn't viable If there's a validator you'd like to see on XDP, tag them below👇
1
10
596
i have a bloomberg terminal for solana
1
61
here you go
Here's my finalized proposal for the resource fee. Although it's reframed a bit as breaking up the base fee into resource and inclusion fee, the mechanism is identical to that in the discussion with an additional base fee adjustment. This version manages to just slightly **reduce** vote costs and oracle updates (and other hyperefficient txns), while increasing the burn for very inefficient transactions. This is of course tunable, but it is thus suitable to be activated before OR after Alpenglow activation. github.com/solana-foundation… TLDR infographic here: burn.cavey.xyz
2
218
i can do weighted pull-ups with 40kg, but i still can’t do a muscle-up. now my elbow is sore 🫠
2
63
The part worth thinking about is why burn the resource fee 100% instead of splitting it 50/50 with the leader, the way the base signature fee is split today. Start with who does the work. The leader builds the block, but every other node has to replay and vote on it, which means re-executing every transaction. The resource cost a tx imposes isn't paid by the leader alone, it's paid by the whole network. So if a fee prices that cost, it makes more sense for the whole cluster to benefit through the burn than for the one node that happened to be leader to pocket half. There's also an incentive problem with splitting it. Leaders already build blocks to maximize their reward, today that's priority fees plus their half of the base fee. Give them a cut of the resource fee too and you add a new dimension to that: they'd earn more from how many CUs get consumed and from any traffic that pays, including wasteful traffic. At the margin you'd be paying block producers to fill blocks with heavy junk. Full burn removes that lever, the leader has no reason to care how many resources a tx burns through if they aren't paid for it. And this takes nothing from validators. base/2 priority fees stay exactly as today. The whole debate is just whether this one new fee stream gets burned or captured. On scale, even rough numbers point the same way. Using the requested-CU data posted in the thread (~16-18T CUs/day) at 0.1 lamport, the burn lands somewhere around 1,500-2,200 SOL/day at current usage. A 50/50 split hands maybe ~750-1,100 SOL/day to the entire validator set combined. Per validator, that's noise. Now weigh that against the base it's competing with. Very roughly, hundreds of millions of SOL are staked earning mid-single-digit APY, which is on the order of tens of millions of SOL/yr in rewards across validators and delegators. A validator's reward scales with SOL price across that whole base. I'm not claiming this burn moves price by any specific amount, but you don't need much, even a small improvement in tokenomics is worth more across that staked base than splitting a few hundred SOL/day, and it's shared instead of fought over. That's the trade: capture a tiny per-validator stream now, or burn it and let any value accrue to the whole staked base. The second is bigger in expectation and it's positive sum. Win for Solana.
3
203
Leafar 🦀 retweeted
sometimes less truly is more.. Agave v4.2 reduces the number of repairs requested while lowering block delivery times
8
20
142
23,005