Go to file
Martin Holst Swende a4b70f9ee9 core/vm: less allocations for various call variants (#21222)
* core/vm/runtime/tests: add more benchmarks

* core/vm: initial work on improving alloc count for calls to precompiles

name                                  old time/op    new time/op    delta
SimpleLoop/identity-precompile-10M-6     117ms ±75%      43ms ± 1%  -63.09%  (p=0.008 n=5+5)
SimpleLoop/loop-10M-6                   79.6ms ± 4%    70.5ms ± 1%  -11.42%  (p=0.008 n=5+5)

name                                  old alloc/op   new alloc/op   delta
SimpleLoop/identity-precompile-10M-6    24.4MB ± 0%     4.9MB ± 0%  -79.94%  (p=0.008 n=5+5)
SimpleLoop/loop-10M-6                   13.2kB ± 0%    13.2kB ± 0%     ~     (p=0.357 n=5+5)

name                                  old allocs/op  new allocs/op  delta
SimpleLoop/identity-precompile-10M-6      382k ± 0%      153k ± 0%  -59.99%  (p=0.000 n=5+4)
SimpleLoop/loop-10M-6                     40.0 ± 0%      40.0 ± 0%     ~     (all equal)

* core/vm: don't allocate big.int for touch

name                                  old time/op    new time/op    delta
SimpleLoop/identity-precompile-10M-6    43.3ms ± 1%    42.4ms ± 7%     ~     (p=0.151 n=5+5)
SimpleLoop/loop-10M-6                   70.5ms ± 1%    76.7ms ± 1%   +8.67%  (p=0.008 n=5+5)

name                                  old alloc/op   new alloc/op   delta
SimpleLoop/identity-precompile-10M-6    4.90MB ± 0%    2.46MB ± 0%  -49.83%  (p=0.008 n=5+5)
SimpleLoop/loop-10M-6                   13.2kB ± 0%    13.2kB ± 1%     ~     (p=0.571 n=5+5)

name                                  old allocs/op  new allocs/op  delta
SimpleLoop/identity-precompile-10M-6      153k ± 0%       76k ± 0%  -49.98%  (p=0.029 n=4+4)
SimpleLoop/loop-10M-6                     40.0 ± 0%      40.0 ± 0%     ~     (all equal)

* core/vm: reduce allocs in staticcall

name                                  old time/op    new time/op    delta
SimpleLoop/identity-precompile-10M-6    42.4ms ± 7%    37.5ms ± 6%  -11.68%  (p=0.008 n=5+5)
SimpleLoop/loop-10M-6                   76.7ms ± 1%    69.1ms ± 1%   -9.82%  (p=0.008 n=5+5)

name                                  old alloc/op   new alloc/op   delta
SimpleLoop/identity-precompile-10M-6    2.46MB ± 0%    0.02MB ± 0%  -99.35%  (p=0.008 n=5+5)
SimpleLoop/loop-10M-6                   13.2kB ± 1%    13.2kB ± 0%     ~     (p=0.143 n=5+5)

name                                  old allocs/op  new allocs/op  delta
SimpleLoop/identity-precompile-10M-6     76.4k ± 0%      0.1k ± 0%     ~     (p=0.079 n=4+5)
SimpleLoop/loop-10M-6                     40.0 ± 0%      40.0 ± 0%     ~     (all equal)

* trie: better use of hasher keccakState

* core/state/statedb: reduce allocations in getDeletedStateObject

* core/vm: reduce allocations in all call derivates

* core/vm: reduce allocations in call variants

- Make returnstack `uint32`
- Use a `sync.Pool` of `stack`s

* core/vm: fix tests

* core/vm: goimports

* core/vm: tracer fix + staticcall gas fix

* core/vm: add back snapshot to staticcall

* core/vm: review concerns + make returnstack pooled + enable returndata in traces

* core/vm: fix some test tracer method signatures

* core/vm: run gencodec, minor comment polish

Co-authored-by: Péter Szilágyi <peterke@gmail.com>
# Conflicts:
#	core/state/statedb.go
#	core/vm/contracts_test.go
#	core/vm/evm.go
#	core/vm/instructions.go
#	core/vm/interpreter.go
#	core/vm/logger.go
#	core/vm/logger_json.go
#	core/vm/logger_test.go
#	core/vm/runtime/runtime_test.go
#	core/vm/stack/stack.go
#	eth/tracers/tracer.go
#	eth/tracers/tracer_test.go
#	trie/secure_trie.go
2020-08-07 13:46:26 +02:00
.buildkite Nightly tests (#444) 2020-04-12 19:37:15 +01:00
.circleci [wip] Cache z3 build on CI, up golangci-lint to v1.27.0 (#615) 2020-06-04 09:54:17 +01:00
.github .github: Change Code Owners (#21326) 2020-08-07 13:17:56 +02:00
.golangci Use MOV instead of VMOV in fnvHash16AVX2 to peacify asmdecl (#699) 2020-06-30 11:23:28 +01:00
accounts accounts/external: remove dependency on internal/ethapi (#21319) 2020-08-07 12:55:18 +02:00
build Fixup for TestSendTransactions65 (#838) 2020-07-31 10:44:42 +02:00
cmd core/vm: less allocations for various call variants (#21222) 2020-08-07 13:46:26 +02:00
common common/math: use math/bits intrinsics for Safe* (#21316) 2020-08-07 12:52:46 +02:00
consensus les: historical data garbage collection (#19570) 2020-08-07 13:16:46 +02:00
console ethapi: don't crash when keystore-specific methods are called but external signer used (#21279) 2020-08-07 12:27:36 +02:00
contracts/checkpointoracle accounts/abi: move U256Bytes to common/math (#21020) 2020-05-20 15:26:22 +03:00
core core/vm: less allocations for various call variants (#21222) 2020-08-07 13:46:26 +02:00
crypto core: types: less allocations when hashing and tx handling (#21265) 2020-08-07 11:46:33 +02:00
debug-web-ui Docker compose (#841) 2020-08-01 09:39:04 +02:00
design auto-format code by prettier (similar to gofmt) (#405) 2020-03-25 12:45:21 +07:00
docs docs: add a piechart with stages 2020-08-05 10:29:15 +02:00
eth core/vm: less allocations for various call variants (#21222) 2020-08-07 13:46:26 +02:00
ethclient Speed up GenerateChain by using intermediate hashes (#736) 2020-07-10 22:37:34 +01:00
ethdb drop badger support (#869) 2020-08-05 16:33:45 +01:00
ethstats ethstats: use timer instead of time.sleep (#20924) 2020-08-06 14:08:02 +02:00
event event, whisper/whisperv6: use defer where possible (#20940) 2020-05-20 15:26:22 +03:00
graphql les: historical data garbage collection (#19570) 2020-08-07 13:16:46 +02:00
internal cmd/clef: change --rpcport to --http.port and update flags in docs (#21318) 2020-08-07 13:23:29 +02:00
log all: fix typos in comments (#21118) 2020-06-15 19:38:13 +03:00
metrics cmd/geth: allow configuring metrics HTTP server on separate endpoint (#21290) 2020-08-07 12:33:14 +02:00
migrations Migrations: use stage name as db key (#868) 2020-08-05 17:13:35 +07:00
miner eth/downloader: fix spuriously failing tests (#21149) 2020-08-07 11:15:45 +02:00
node cmd, node: dump empty value config (#21296) 2020-08-07 12:47:04 +02:00
p2p p2p/discover: require table nodes to have an IP (#21330) 2020-08-07 13:18:23 +02:00
params les: historical data garbage collection (#19570) 2020-08-07 13:16:46 +02:00
rlp rlp: reduce allocations for big.Int and byte array encoding (#21291) 2020-08-07 12:46:45 +02:00
rpc post-rebase fixups 2020-06-15 19:38:54 +03:00
semantics EVM semantics - writeup (#450) 2020-04-14 13:49:38 +01:00
signer post-rebase fixups 2020-06-15 19:38:54 +03:00
tests cmd/evm: add state transition tool for testing (#20958) 2020-08-07 11:38:07 +02:00
trie core: types: less allocations when hashing and tx handling (#21265) 2020-08-07 11:46:33 +02:00
visual Continue comparison of genesis block with geth, expand long values (#223) 2019-12-06 12:03:12 +00:00
.dockerignore Add support of geth on hostmachine (#437) 2020-04-11 08:22:23 +01:00
.gitattributes .gitattributes: enable solidity highlighting on github (#16425) 2018-04-03 15:21:24 +02:00
.gitignore Docker compose (#841) 2020-08-01 09:39:04 +02:00
.gitmodules Semantics: Integrate Z3 into the build (#370) 2020-03-06 08:54:21 +00:00
.golangci.yml build: upgrade to golangci lint v1.27.0 (#21127) 2020-06-15 19:38:13 +03:00
.mailmap all: update license information (#16089) 2018-02-14 13:49:11 +01:00
.travis.yml eth/downloader: more context in errors (#21067) 2020-06-15 19:38:13 +03:00
appveyor.yml geth 1.9.13 (#469) 2020-04-19 18:31:47 +01:00
AUTHORS build: deduplicate same authors with different casing 2019-07-22 12:31:11 +03:00
circle.yml circleci: enable docker based hive testing 2016-07-15 16:07:34 +03:00
COPYING all: update license information 2015-07-07 14:12:44 +02:00
COPYING.LESSER all: update license information 2015-07-07 14:12:44 +02:00
docker-compose.yml Disable ipc RPC (#853) 2020-08-02 12:49:01 +01:00
Dockerfile Docker compose (#841) 2020-08-01 09:39:04 +02:00
Dockerfile.alltools Docker compose (#841) 2020-08-01 09:39:04 +02:00
fuzzbuzz.yaml eth: rework tx fetcher to use O(1) ops + manage network requests 2020-02-27 17:21:20 +03:00
go.mod go.mod: upgrade to github.com/golang/snappy with arm64 asm (#21304) 2020-08-07 12:50:25 +02:00
go.sum drop badger support (#869) 2020-08-05 16:33:45 +01:00
interfaces.go [GC] uint256 rather than big.Int in Transaction (#614) 2020-06-04 08:43:08 +01:00
Makefile drop badger support (#869) 2020-08-05 16:33:45 +01:00
nightly.sh Nightly tests (#444) 2020-04-12 19:37:15 +01:00
README.geth.md New readme.md (#827) 2020-07-30 12:04:46 +01:00
README.md resident_memory_docs (#864) 2020-08-04 09:03:59 +01:00
RELEASE_INSTRUCTIONS.md Jumpdest skip optimisation (#851) 2020-08-01 17:56:57 +01:00
SECURITY.md SECURITY.md: create security policy (#19666) 2019-06-06 14:40:52 +02:00
to-merge.txt trie: quell linter in commiter.go (#21329) 2020-08-07 13:24:03 +02:00
UPGRADE_INFO.md prepare for merging 2020-02-27 17:20:35 +03:00

Turbo-Geth

Turbo-Geth is a fork of Go-Ethereum with focus on performance. CircleCI

Table of contents

NB! In-depth links are marked by the microscope sign (🔬)

Disclaimer: this software is currenly a tech preview. We will do our best to keep it stable and make no breaking changes but we don't guarantee anything. Things can and will break.

The current version is currently based on Go-Ethereum 1.9.15.

System Requirements

About 830 GB of free disk storage (630 GB state storage, 200GB temp files)

16 or 32 GB of RAM is recommended

🔬 more info on disk storage is here here)

Usage

> git clone --recurse-submodules -j8 https://github.com/ledgerwatch/turbo-geth.git && cd turbo-geth
> make tg
> ./build/bin/tg

Key features

🔬 See more detailed overview of functionality and current limitations. It is being updated on recurring basis.

More Efficient State Storage

Flat KV storage. Turbo-Geth uses a key-value database and storing accounts and storage in a simple way.

🔬 See our detailed DB walkthrough here.

Preprocessing. For some operations, turbo-geth uses temporary files to preprocess data before inserting it into the main DB. That reduces write amplification and DB inserts sometimes are orders of magnitude quicker.

Plain state.

Single accounts/state trie. Turbo-Geth uses a single Merkle trie for both accounts and the storage.

Faster Initial Sync

Turbo-Geth uses a rearchitected full sync algorithm from Go-Ethereum that is split into "stages".

🔬 See more detailed explanation in the Staged Sync Readme

It uses the same network primitives and is compatible with regular go-ethereum nodes that are using full sync, you do not need any special sync capabilities for turbo-geth to sync.

When reimagining the full sync, we focused on batching data together and minimize DB overwrites. That makes it possible to sync Ethereum mainnet in under 2 days if you have a fast enough network connection and an SSD drive.

Examples of stages are:

  • Downloading headers;

  • Downloading block bodies;

  • Executing blocks;

  • Validating root hashes and building intermediate hashes for the state Merkle trie;

  • And more...

JSON-RPC daemon

In turbo-geth RPC calls are extracted out of the main binary into a separate daemon. This daemon can use both local or remote DBs. That means, that this RPC daemon doesn't have to be running on the same machine as the main turbo-geth binary or it can run from a snapshot of a database for read-only calls.

🔬 See RPC-Daemon docs

For local DB

> make rpcdaemon
> ./build/bin/rpcdaemon --chaindata ~/Library/TurboGeth/tg/chaindata --http.api=eth,debug

For remote DB

Run turbo-geth in one terminal window

> ./build/bin/tg --private.api.addr=localhost:9090

Run RPC daemon

> ./build/bin/rpcdaemon --private.api.addr=localhost:9090

Supported JSON-RPC calls (eth, debug):

eth_call
eth_getBlockByHash
eth_getBlock
eth_blockNumber
eth_getBalance
eth_getLogs
eth_estimateGas
debug_storageRangeAt
debug_traceTransaction
debug_accountRange
debug_getModifiedAccountsByNumber
debug_getModifiedAccountsByHash

REST API Daemon

Apart from JSON-RPC daemon, Turbo-Geth also contains REST API daemon. It uses turbo-geth remote DB functionality.

🔬 See REST API docs

Run turbo-geth in one terminal window

> ./build/bin/tg --private.api.addr=localhost:9090

Run REST daemon

> make restapi
> ./build/bin/restapi --private.api.addr=localhost:9090

This API is very limited at the moment too:

GET /api/v1/accounts/<accountAddress>
GET /api/v1/storage/?prefix=PREFIX

Or run all components by docker-compose

Next command starts: turbo-geth on port 30303, rpcdaemon 8545, restapi 8080, debug-web-ui 3001, prometheus 9090, grafana 3000

docker-compose build
XDG_DATA_HOME=/preferred/data/folder docker-compose up

Getting in touch

Turbo-Geth Discord Server

The main discussions are happening on our Discord server. To get an invite, send an email to tg [at] torquem.ch with your name, occupation, a brief explanation of why you want to join the Discord, and how you heard about Turbo-Geth.

Reporting security issues/concerns

Send an email to security [at] torquem.ch.

Team

Core contributors:

Thanks to:

  • All contributors of Turbo-Geth

  • All contributors of Go-Ethereum

  • Our special respect and graditude is to the core team of Go-Ethereum. Keep up the great job!

Happy testing! 🥤

Known issues

htop shows incorrect memory usage

TurboGeth's internal DB (LMDB) using MemoryMap - when OS does manage all read, write, cache operations instead of Application (linux, windows)

htop on column res shows memory of "App + OS used to hold page cache for given App", but it's not informative, because if htop says that app using 90% of memory you still can run 3 more instances of app on the same machine - because most of that 90% is "OS pages cache".
OS automatically free this cache any time it needs memory. Smaller "page cache size" may not impact performance of TurboGeth at all.

Next tools show correct memory usage of TurboGeth:

  • vmmap -summary PID | grep -i "Physical footprint". Without grep you can see details - section MALLOC ZONE column Resident Size shows App memory usage, section REGION TYPE column Resident Size shows OS pages cache size.
  • Prometheus dashboard shows memory of Go app without OS pages cache (make prometheus, open in browser localhost:3000, credentials admin/admin)
  • cat /proc/<PID>/smaps

TurboGeth uses ~4Gb of RAM during genesis sync and < 1Gb during normal work. OS pages cache can utilize unlimited amount of memory.

Warning: Multiple instances of TG on same machine will touch Disk concurrently, it impacts performance - one of main TG optimisations: "reduce Disk random access". "Blocks Execution stage" still does much random reads - this is reason why it's slowest stage. We do not recommend run multiple genesis syncs on same Disk. If genesis sync passed, then it's fine to run multiple TG on same Disk.