a069738c20
* update shared/params * update eth2-types deps * update protobufs * update shared/* * fix testutil/state * update beacon-chain/state * update beacon-chain/db * update tests * fix test * update beacon-chain/core * update beacon-chain/blockchain * update beacon-chain/cache * beacon-chain/forkchoice * update beacon-chain/operations * update beacon-chain/p2p * update beacon-chain/rpc * update sync/initial-sync * update deps * update deps * go fmt * update beacon-chain/sync * update endtoend/ * bazel build //beacon-chain - works w/o issues * update slasher code * udpate tools/ * update validator/ * update fastssz * fix build * fix test building * update tests * update ethereumapis deps * fix tests * update state/stategen * fix build * fix test * add FarFutureSlot * go imports * Radek's suggestions * Ivan's suggestions * type conversions * Nishant's suggestions * add more tests to rpc_send_request * fix test * clean up * fix conflicts Co-authored-by: prylabs-bulldozer[bot] <58059840+prylabs-bulldozer[bot]@users.noreply.github.com> Co-authored-by: nisdas <nishdas93@gmail.com> |
||
|---|---|---|
| .. | ||
| benchmark_files | ||
| BUILD.bazel | ||
| pregen_test.go | ||
| pregen.go | ||
| README.md | ||
Benchmarks for Prysm State Transition
This package contains the functionality needed for benchmarking Prysms state transitions, this includes its block processing (with and without caching) and epoch processing functions. There is also a benchmark for HashTreeRoot on a large beacon state.
Benchmark Configuration
The following configs are in config.go:
ValidatorCount: Sets the amount of active validators to perform the benchmarks with. Default is 16384.AttestationsPerEpoch: Sets the amount of attestations per epoch for the benchmarks to perform with, this affects the amount of attestations in a full block and the amount of attestations per epoch in the state for theProcessEpochandHashTreeRootbenchmark. Default is 128.
Generating new SSZ files
Due to the sheer size of the benchmarking configurations (16384 validators), the files used for benchmarking are pregenerated so there's no wasted computations on generating a genesis state with 16384 validators. This should only be needed if there is a breaking spec change and the tests fail from SSZ issues.
To generate new files to use for benchmarking, run the below command in the root of Prysm.
bazel run //tools/benchmark-files-gen -- --output-dir $PRYSMPATH/shared/benchutil/benchmark_files/ --overwrite
Running the benchmarks
To run the ExecuteStateTransition benchmark:
bazel test //beacon-chain/core/state:go_default_test --test_filter=BenchmarkExecuteStateTransition_FullBlock --test_arg=-test.bench=BenchmarkExecuteStateTransition_FullBlock
To run the ExecuteStateTransition (with cache) benchmark:
bazel test //beacon-chain/core/state:go_default_test --test_filter=BenchmarkExecuteStateTransition_WithCache --test_arg=-test.bench=BenchmarkExecuteStateTransition_WithCache
To run the ProcessEpoch benchmark:
bazel test //beacon-chain/core/state:go_default_test --test_filter=BenchmarkProcessEpoch_2FullEpochs --test_arg=-test.bench=BenchmarkProcessEpoch_2FullEpochs
To run the HashTreeRoot benchmark:
bazel test //beacon-chain/core/state:go_default_test --test_filter=BenchmarkHashTreeRoot_FullState --test_arg=-test.bench=BenchmarkHashTreeRoot_FullState
To run the HashTreeRootState benchmark:
bazel test //beacon-chain/core/state:go_default_test --test_filter=BenchmarkHashTreeRootState_FullState --test_arg=-test.bench=BenchmarkHashTreeRootState_FullState
Extra flags needed to benchmark properly:
--nocache_test_results --test_arg=-test.v --test_timeout=2000 --test_arg=-test.cpuprofile=/tmp/cpu.profile --test_arg=-test.memprofile=/tmp/mem.profile --test_output=streamed
Current Results as of January 2020
BenchmarkExecuteStateTransition_FullBlock-4 20 2031438030 ns/op
BenchmarkExecuteStateTransition_WithCache-4 20 1857290454 ns/op
BenchmarkHashTreeRoot_FullState-4 50 297655834 ns/op
BenchmarkHashTreeRootState_FullState-4 50 155535883 ns/op