Pinned
Alessandro Decina
10.3K posts
- Writing C for the first time after exclusively writing Rust for many years is a bittersweet experience. It's like when every few years I visit my home town up in the mountains in the middle of nowhere. Nature is amazing. Life is so calm. Everything seems so simple. But
- This was an anza dev node during its leader slots a couple of minutes ago. TPU is at 7% load under current traffic. We have 93% free capacity. TPU is ready for 1M TPS.
- Random thoughts now that things have settled a bit (going to regret saying this in 3, 2, 1...) What happened the past few days is amazing, and I'm still buzzing. We did great, despite what some of the usual larpers are saying (more on that below). Agave ingested traffic at peak
- Remember when I shared this almost exactly a month ago? x.com/alessandrod/st… It's fixed. For the first time ever, we're actually building blocks in parallel. I've highlighted a block our dev validator produced in _under 110 milliseconds_. Do you know what this means? Do you
Replying to @alessandrodNext let's look at the threads you probably most care about: the ones that process the user txs that end up in the blocks. At the very start of the first block, all the workers execute transactions in _parallel_. Parallel good. Unfortunately the work quickly becomes - we do 100k TPS and feel deep dissatisfaction you do 19k TPS with creative crypto accounting across 50 chains and celebrate we are not the sameEthereum L2s did 19,000 TPS in the last 24 hours
- ethereum bag holders trying every move to divert attention from the fact that their chain does 10 TPS
00:00 - Traffic spiked about 6x from baseline. I slept in. Woke up. Checked slack. Nothing out of ordinary. Say the word congestion one more time
Over the last hour, Solana's network demonstrated resilience under extreme demand: - Raw transactions spiked to 6-10k per second - Utilization reached up near 60 CUs per block - Median transaction fees remained low - rough agave perf timeline: 2.2: - almost completely mmap free - tx execution nearly I/O free with upcoming WAL fix and --disable-accounts-disk-index - 2x block sizes 2.3: - tx execution I/O free - isolated arenas for banking and replay - isolated banking and replay threads
Hyper-parallel shared state is the purest form of web3 interoperability & value. Only on Aptos.- Took a couple weeks longer than expected because I found other bugs along the way, but here's some numbers on the XDP work I've been doing. tldr: turbine can now retransmit much faster and is not a bottleneck anymore, and does so using a fraction of the CPU we use today. As I
- THROWN TO THE WOLVES
00:00 - if you're a cracked rust/perf engineer at microsoft, send DM we're fully remoteMICROSOFT: EMPLOYEES TO BE REQUIRED TO BE IN OFFICE 3 DAYS/WEEK
