This PR addresses one of the biggest performance issue with binary
tries: storing each internal node individually bloats the index, the
disk, and triggers a lot of write amplifications. To fix this issue,
this PR serializes groups of nodes together.
Because we are still looking for the ideal group size, the "depth" of
the group tree is made a parameter, but that will be removed in the
future, once the perfect size is known.
This is a rebase of #33658
---------
Co-authored-by: Copilot <copilot@github.com>
## Summary
Replace the `BinaryNode` interface with `NodeRef uint32` indices into
typed arena pools, eliminating GC-scanned pointers from binary trie
nodes.
Inspired by [fjl's
observation](https://github.com/ethereum/go-ethereum/pull/34034#issuecomment-4075176446):
> *"if the binary trie produces such a large graph, it should probably
be changed so that the trie node type does not contain pointers. The
runtime does not scan objects that do not contain pointers, so it can
really help with the performance to build it this way."*
### The problem
CPU profiling of the binary trie (EIP-7864) showed **44% of CPU time in
garbage collection**. Each `InternalNode` held two `BinaryNode`
interface values (2 pointer-words each), and the GC scanned every one.
With ~25K `InternalNode`s in memory during block processing, this
created enormous GC pressure.
### The solution
`NodeRef` is a compact `uint32` (2-bit kind tag + 30-bit pool index).
`NodeStore` manages chunked typed pools per node kind:
- **InternalNode pool**: ZERO Go pointers (children are `NodeRef`, hash
is `[32]byte`) → noscan spans
- **HashedNode pool**: ZERO Go pointers → noscan spans
- **StemNode pool**: retains `Values [][]byte` (matching existing
format)
The serialization format is unchanged — flat InternalNode
`[type][leftHash][rightHash]` = 65 bytes.
## Benchmark: Apple M4 Pro (`--benchtime=10s --count=3`, on top of
#34021)
| Metric | Baseline | Arena | Delta |
|--------|----------|-------|-------|
| Approve (Mgas/s) | 374 | 382 | **+2.1%** |
| BalanceOf (Mgas/s) | 885 | 901 | **+1.8%** |
| Approve allocs/op | 775K | **607K** | **-21.7%** |
| BalanceOf allocs/op | 265K | **228K** | **-14.0%** |
## Benchmark: AMD EPYC 48-core (50GB state, execution-specs ERC-20, on
top of #34021 + #34032)
| Benchmark | Baseline | Arena | Delta |
|-----------|----------|-------|-------|
| erc20_approve (write) | 22.4 Mgas/s | **27.0 Mgas/s** | **+20.5%** |
| mixed_sload_sstore | 62.9 Mgas/s | **97.3 Mgas/s** | **+54.7%** |
| erc20_balanceof (read) | 180.8 Mgas/s | 167.6 Mgas/s | -7.3% (cold
cache variance) |
The arena benefit scales with heap size — the EPYC (larger heap, more GC
pressure) shows much larger gains than the M4 Pro (efficient unified
memory). The mixed workload baseline was unstable (62.9 vs 16.3 Mgas/s
between runs due to GC-induced throughput collapse); the arena
eliminates this entirely (95-97 Mgas/s, stable).
## Dependencies
Benchmarked with #34021 (H01 N+1 fix) + #34032 (R14 parallel hashing).
No code dependency — applies independently to master.
All test suites pass (`trie/bintrie` with `-race`, `core/state`,
`triedb/pathdb`, `cmd/geth`).
---------
Co-authored-by: Guillaume Ballet <3272758+gballet@users.noreply.github.com>
## Problem
`BinaryTrie.Commit` unconditionally walked every resolved in-memory node
and flushed it into the `NodeSet`, producing one Pebble write per
resolved internal + stem node on every block — even when the node's
on-disk blob was bitwise identical to the previous commit. On a warm
400M-state workload this meant tens of thousands of redundant 65-byte
writes per block, compounding Pebble compaction pressure on every
commit.
The existing `mustRecompute` flag tracks *hash* staleness, not
*disk-blob* staleness: after `Hash()` completes, `mustRecompute` is
cleared even though the fresh blob has not been persisted. It is
therefore insufficient for a skip-flush optimization.
## Fix
Mirror the MPT committer pattern (`trie/committer.go:51-56`) by adding a
`dirty` flag on `InternalNode` and `StemNode` with the semantics *the
on-disk blob is stale*. The flag is:
- set to `true` wherever the node is created or structurally modified
(the same call sites that already set `mustRecompute = true`);
- set to `false` only after the node has been passed to the `flushfn`
inside `CollectNodes`;
- left `false` on nodes produced by `DeserializeNodeWithHash`, matching
the *loaded from disk, already persisted* semantics.
`CollectNodes` short-circuits on `!dirty` subtrees. The propagation
invariant (an ancestor of any dirty node is itself dirty) is already
maintained by the existing `InsertValuesAtStem` / `Insert` paths, which
now mirror every `mustRecompute = true` setter with a `dirty = true`
setter.
## Benchmark
New `BenchmarkCollectNodes_SparseWrite` measures commit cost when only
one leaf changes between blocks — the common case for state updates.
10,000-stem trie, one-leaf modification + Commit per iteration, Apple M4
Pro:
| | before | after | delta |
|---|---|---|---|
| time / op | 12,653,000 ns | 7,336 ns | **~1,725×** |
| bytes / op | 107,224,740 B | 37,774 B | **~2,839×** |
| allocs / op | 80,953 | 134 | **~604×** |
End-to-end impact on a real workload depends on the
resolved-footprint-to-dirty-path ratio; the new
`TestBinaryTrieCommitIncremental` provides a structural regression guard
(asserts that a Commit following a single-leaf modification flushes a
root-to-leaf path, not the whole tree).
---
Found all of this stuff while bloating my #34706 DB to make some
benchmarks. And saw we were spending A LOT OF TIME on hashing.
Hope this helps the perf a bit. Will rebase the flat-state PR on top of
this once merged.
This is an optimization that existed for verkle and the MPT, but that
got dropped during the rebase.
Mark the nodes that were modified as needing recomputation, and skip the
hash computation if this is not needed. Otherwise, the whole tree is
hashed, which kills performance.
Implement the binary tree as specified in [eip-7864](https://eips.ethereum.org/EIPS/eip-7864).
This will gradually replace verkle trees in the codebase. This is only
running the tests and will not be executed in production, but will help
me rebase some of my work, so that it doesn't bitrot as much.
---------
Signed-off-by: Guillaume Ballet
Co-authored-by: Parithosh Jayanthi <parithosh.jayanthi@ethereum.org>
Co-authored-by: rjl493456442 <garyrong0905@gmail.com>
