XDR opt experiment#5300

Draft

dmkozh wants to merge 50 commits into

stellar:masterfrom

dmkozh:ai_opt_p27

Commits on May 28, 2026

budget opt step 1
dmkozh
committed
rollback env, update benchmark config
dmkozh
committed
disable test meta
dmkozh
committed
Streaming SHA256 for InvokeHostFunction success hash

SirTyson
authored and
dmkozh
committed
Parallelize TxFrame creation and transaction validation

authored and
dmkozh
committed
validate txs in parallel, small improvement on some tests (?)
dmkozh
committed
Cache XDR size in InMemorySorobanState entries

authored and
dmkozh
committed
Parallelize in-memory state update with bucket list operations

SirTyson
authored and
dmkozh
committed
Parallel pre-apply 5-20ms
dmkozh
committed
profile flag for bench matrix
dmkozh
committed
Cache ledger info
dmkozh
committed
add config flag for ledger close worker threads
dmkozh
committed
Detailed apply stage breakdown
dmkozh
committed
Optimize `rescope` using move.
dmkozh
committed
add tracy support to bench matrix
dmkozh
committed
Switch SHA256 from libsodium (pure C) to OpenSSL (SHA-NI hardware accel)

authored and
dmkozh
committed
Parallelize InMemoryIndex construction with bucket put loop (saves ~25ms/ledger)

authored and
dmkozh
committed
perf: eliminate per-tx child LTX in fee processing (+19.2% TPS)

Garand Tyson
authored and
dmkozh
committed
perf: track entry existence in ParallelApplyEntry to skip SHA256 lookups

Garand Tyson
authored and
dmkozh
committed
perf: move semantics in commitChangesToLedgerTxn to avoid XDR copies

authored and
dmkozh
committed
perf: pre-load Soroban RO entries + processFeesSeqNums optimizations

Garand Tyson
authored and
dmkozh
committed
Optimize recordStorageChanges.
dmkozh
committed
perf: reserve parallel apply container capacity to eliminate rehashing

Garand Tyson
authored and
dmkozh
committed
Remove extra lookup from upsert
dmkozh
committed
update scenarios
dmkozh
committed
More robust path handling in apply load matrix script
dmkozh
committed
perf: avoid building 128K-entry modifiedKeys set for eviction scan

authored and
dmkozh
committed
Shard verifySig cache to reduce mutex contention (7680→8896 TPS, +15.8%)

Garand Tyson
authored and
dmkozh
committed
perf: indirect sort in convertToBucketEntry (+2.8% TPS)

Garand Tyson
authored and
dmkozh
committed
perf: skip invariant delta when no invariants enabled (+8.0% TPS)

Garand Tyson
authored and
dmkozh
committed
Cache LedgerKey hash in parallel apply data structures - ~-5ms
dmkozh
committed
Manual txset building instrumentation
dmkozh
committed
storage opt
dmkozh
committed
revert host module to p26
dmkozh
committed
format
dmkozh
committed
fix a bug - in-memory state update shouldn't be conditioned on protocol version
dmkozh
committed
Adapt to upstream LedgerSnapshot/ApplyLedgerStateSnapshot rename
dmkozh
committed
Use ApplyLedgerView::asImmutableView for parallel pre-apply
dmkozh
committed
Add missing inclusion-fee check to parallel TxFrame creation paths
dmkozh
committed
xdr opt
dmkozh
committed
perf: eliminate avoidable deep copies in bucket merge and TxSetFrame
dmkozh
committed
perf: single-pass writeOne and raw byte output in XDRStream
dmkozh
committed
perf: raw byte passthrough in bucket merge to skip re-serialization
dmkozh
committed
use scratch buffer for XDR encode where applicable
dmkozh
committed
xdrpp xdr optimizations
dmkozh
committed