nimbus-eth2/nbench
Jacek Sieka a086cf01ac
altair fork handling cleanups ()
* fix stack overflow crash in REST/debug/getStateV2
* introduce `ForkyXxx` for generic type matching of `Xxx` across
branches (SomeHashedBeaconState -> ForkyHashedBeaconState et al) -
`Some` is already used for other types of type classes
* consolidate function naming in BeaconChainDB, use some generics
* import `forks.nim` from other spec modules and move `Forked*` helpers
around to resolve circular imports
* remove `ForkedBeaconState`, use `ForkedHashedBeaconState` throughout
(less data shuffling between the types)
* fix several cases of states being stored on stack in tests, causing
random failures on some platforms
* remove reading json support from ncli - this should be ported to the
rest json reading instead (doesn't currently work because stack sizes)
2021-11-05 08:34:34 +01:00
..
platforms cron CI: various fixes () 2021-08-18 14:40:03 +02:00
README.md split state_transition() into slots/block parts and use only block where appropriate () 2021-06-03 11:42:25 +02:00
bench_lab.nim Revive nbench () 2020-05-15 00:54:10 +02:00
foo.nim Nbench - Flexible benchmarking of Nimbus internals () 2019-12-20 17:14:43 +01:00
nbench.nim remove false OnBlockAdded dependency on phase0 HashedBeaconState () 2021-06-21 08:35:24 +00:00
nbench.nim.cfg Nbench - Flexible benchmarking of Nimbus internals () 2019-12-20 17:14:43 +01:00
nbench_spec_scenarios.nim test fixture cleanups () 2021-10-12 13:36:52 +02:00
reports.nim Revive nbench () 2020-05-15 00:54:10 +02:00
scenarios.nim altair fork handling cleanups () 2021-11-05 08:34:34 +01:00

README.md

Nimbus-bench

Nbench is a profiler dedicated to the Nimbus Beacon Chain.

It is built as a domain specific profiler that aims to be as unintrusive as possible while providing complementary reports to dedicated tools like perf, Apple Instruments or Intel Vtune that allows you to dive deep down to a specific line or assembly instructions.

In particular, those tools cannot tell you that your cryptographic subsystem or your parsing routines or your random number generation should be revisited, may sample at to high a resolution (millisecond) instead of per-function statistics, and are much less useful without debugging symbols which requires a lot of space. I.e. perf and other generic profiler tools give you the laser-thin focused pictures while nbench strives to give you the big picture.

Features

  • by default nbench will collect the number of calls and time spent in each function.
  • like ncli or nfuzz, you can provide nbench isolated scenarios in SSZ format to analyze Nimbus behaviour.

Usage

nim c -d:const_preset=mainnet -d:nbench -d:release -o:build/nbench nbench/nbench.nim
export SCENARIOS=vendor/nim-eth2-scenarios/tests-v0.11.1/mainnet/phase0

# Full state transition
build/nbench cmdFullStateTransition -d="${SCENARIOS}"/sanity/blocks/pyspec_tests/voluntary_exit/ -q=2

# Slot processing
build/nbench cmdSlotProcessing -d="${SCENARIOS}"/sanity/slots/pyspec_tests/slots_1

# Justification-Finalisation
build/nbench cmdEpochProcessing --epochProcessingCat=catJustificationFinalization -d="${SCENARIOS}"/epoch_processing/justification_and_finalization/pyspec_tests/234_ok_support/

# Registry updates
build/nbench cmdEpochProcessing --epochProcessingCat=catRegistryUpdates -d="${SCENARIOS}"/epoch_processing/registry_updates/pyspec_tests/activation_queue_efficiency/

# Slashings
build/nbench cmdEpochProcessing --epochProcessingCat=catSlashings -d="${SCENARIOS}"/epoch_processing/slashings/pyspec_tests/max_penalties/

# Block header processing
build/nbench cmdBlockProcessing --blockProcessingCat=catBlockHeader -d="${SCENARIOS}"/operations/block_header/pyspec_tests/proposer_slashed/

# Proposer slashing
build/nbench cmdBlockProcessing --blockProcessingCat=catProposerSlashings -d="${SCENARIOS}"/operations/proposer_slashing/pyspec_tests/invalid_proposer_index/

# Attester slashing
build/nbench cmdBlockProcessing --blockProcessingCat=catAttesterSlashings -d="${SCENARIOS}"/operations/attester_slashing/pyspec_tests/success_surround/

# Attestation processing
build/nbench cmdBlockProcessing --blockProcessingCat=catAttestations -d="${SCENARIOS}"/operations/attestation/pyspec_tests/success_multi_proposer_index_iterations/

# Deposit processing
build/nbench cmdBlockProcessing --blockProcessingCat=catDeposits -d="${SCENARIOS}"/operations/deposit/pyspec_tests/new_deposit_max/

# Voluntary exit
build/nbench cmdBlockProcessing --blockProcessingCat=catVoluntaryExits -d="${SCENARIOS}"/operations/voluntary_exit/pyspec_tests/validator_exit_in_future/

Running the whole Eth2.0 specs test suite

Warning: this is a proof-of-concept, there is a slight degree of interleaving in output. Furthermore benchmarks are run in parallel and might interfere which each other.

nim c -d:const_preset=mainnet -d:nbench -d:release -o:build/nbench nbench/nbench.nim
nim c -o:build/nbench_tests nbench/nbench_spec_scenarios.nim
build/nbench_tests --nbench=build/nbench --tests=vendor/nim-eth2-scenarios/tests-v0.11.1/mainnet/

TODO Reporting

  • Dumping as CSV files also for archival, perf regression suite and/or data mining.
  • Piggybacking on eth-metrics and can report over Prometheus or StatsD.
  • you can augment it via label pragmas that can be applied file-wide to tag "cryptography", "block_transition", "database" to have a global view of the system.