1/ The Eclipse Performance Thesis. Eclipse is building the Giga Scale Virtual Machine (GSVM), a new SVM client. A software-hardware co-designed, cross-layer optimized, GigaCompute blockchain that eclipses all scalability limits and represents a paradigm shift in performance.

6/ You can read more about this survey of options at:.

5/ In part 2 of the blog, we explain how we achieved this state of the art performance with fine tuning snapshots frequency, subtree roots count, prefetching:

4/ Result?. AlDBaran crushes 24M updates/sec with historization, over 48M without, vastly exceeding goals. You can learn more about the performance in our part 1 blog:

3/ The gap?. None sustain the multi-million updates/sec in production contexts needed for true high-TPS blockchains. This shortfall sparked our AlDBaran: Decoupling in-RAM hot-path (Pleiades) from append-only proofs (Hyades), plus SIMD batching, subtree roots buffering, &.

2/ Firewood's compaction-less trie slashes I/O. MerkleDB's batched views enable stability. NOMT's flash-native layout optimizes SSDs. LVMT's algebraic commitments hit O(1) roots. QMDB's twigs unify storage for 2M+ updates/sec. Impressive. but not enough for prod-scale.

1/ Blockchains are bottlenecked by state updates. We surveyed Firewood, MerkleDB, NOMT, LVMT, & QMDB: Huge advances in I/O and proofs, yet ALL fall orders of magnitude short of our 3M updates/sec for true 1M+ TPS. Enter AlDBaran, smashing 48M/sec!

8/ If you're into learning how crypto makes transactions faster and cheaper, or just building cool tech, this pushes limits to 50M+ updates/sec. Check the full article:

7/ Additional gains:. Special memory allocation adds 25%, larger pages add 5-10%. AlDBaran powers high-speed blockchains like Eclipse but works for others too. Next up: Light nodes, trusted RPCs, disaggregated storage. State commitments no longer bottleneck for high-TPS chains.

6/ Prefetching:. Modern computers "guess" what data you'll need next to avoid delays. We arrange data predictably so the computer loads it ahead, boosting speed by 25% on large memory pages. But watch out: Some cloud services disable this for security reasons, so choose hardware.

5/ More subtree roots alleviate Pleiades load. Our 48M benchmark resides at the midpoint in this range, echoing the tune for your topology. It's like optimizing convex hulls in DeFi AMMs, but for Merkle trees: balancing topology, snapshots, and hardware for peak efficiency.

4/ Another tweak:. "Subtree roots" (like branches in a data tree). Instead of recalculating the whole structure every change, we update smaller sections in parallel and finalize at the end. This cuts unnecessary work. Adjusting the number of these branches can vary speed by.

3/ Think of snapshots like saving a game:. Do it too often, and things slow down; less often, and you go faster but with trade-offs. High-speed apps (e.g., trading) might need frequent saves, while simple viewers can skip them. Benchmarks on AWS 96-core systems and M4 Pro show.

2/ At its core, AlDBaran splits "live" data updates (Pleiades, super-fast in memory) from "history" storage (Hyades, handled separately). This separation lets you adjust settings based on your needs. For example, change how often you "save" snapshots for a 2x speed boost—or.

1/ Blockchains need fast, secure ways to update and verify data. In AlDBaran Part 1, we hit 48M updates/sec (20x faster than rivals). Today, we reveal how to tune AlDBaran for even better results:

RT @EclipseFND: Today we unveil the next step in our protocol’s journey and the evolution of the Eclipse Economy. The Eclipse protocol tok….

5/ We wrap up by highlighting some of our current and future work to get around this problem. You can read all the details here:.

4/ We investigate the feasibility of solving the block packing problem with a state-of-the-art Constrained Programming solver and conclude that the problem is intractable at GigaCompute scale.

3/ We found that the problem can be decomposed into two sub-problems: the fee update rule and the block packing problem (which is NP-hard). There is an extensive body of work on the fee update rule, but the block packing problem is relatively understudied.

2/ The SVM has the edge here: local fee markets allow us to scale the gas limit, and static gas (no refunds) prevents a class of spam. But we decided to step it up a notch and expand the multi-dimensional resource pricing of the SVM.

1/ Fee markets are an often overlooked component when discussing performance, but they’re crucial for scalability. The EVM today can not scale beyond a single execution core because it does not support local fee markets. Additionally, blockspace efficiency is greatly reduced