1/ We just broke the state-commitment bottleneck at Eclipse: AlDBaran sustains 48M updates/sec on a 96-core AWS box, accelerating Eclipse's GigaCompute rollup.

2/ Why it matters: 1M TPS -> 3M state updates/sec (~3 keys/tx on Eclipse). Off-the-shelf ADS engines choke below 0.6 M/s, a 5x gap we refused to accept.

3/ Meet AlDBaran: two purpose-built engines for state. 🔹 Pleiades - blazing-fast, in-DRAM updates 🔹 Hyades - async, append-only proof store

4/ Pleiades' five design rules: 1️⃣ DRAM-only execution (no fsync/page-faults) 2️⃣ Thread-sharding (0 locks) 3️⃣ Twig buffering (defer top-tree hashing) 4️⃣ SIMD batching (16 hashes/vector op) 5️⃣ Predictable layout + prefetch (L2 cache hits)

5/ Benchmark highlights: - 48M ups on 96 cores for Pleiades at roughly 1B keys (0.5 M/core ≈ 78 % solo peak) and 40M ups even at 8 B keys - 24M ups with history for Hyades - State roots now sail through a 50 Gbps link.

6/ That’s a 20× leap over QMDB’s 2.3M/s and 30× faster than our internal QMDB tests. Our 3 M/s requirement now consumes < 7 % of capacity, massive headroom.

7/ Hyades runs fully async, ingesting compact 40B proof entries off-path into an append-only log, while full account payloads land in a separate journal. The hot path stays spotless.

8/ You can learn more about the AlDBaran architecture & headline results at: