refactor: [Memory optimization] Block index allocation overhead #1957
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The Gridcoin researcher context in the block index contained these fields: - cpid (16 bytes) - nResearchSubsidy (8 bytes) - nMagnitude (8 bytes) However, non-researcher nodes produce the overwhelming majority of Gridcoin's blockchain. This change allocates memory for researcher context objects only for blocks staked with a CPID. We add 8 bytes for pointers for every block but save 24 bytes for all of the non- CPID blocks that comprise nearly three-quarters of the chain.
Gridcoin's block chain contains millions of blocks, and nodes need to dynamically allocate memory for small objects to index each of these. Prior to this commit, Gridcoin performed one-off allocations for each block index object. The change replaces those allocations with a pool that allocates many block index objects in chunks. The approach improves memory efficiency by decreasing the overhead of library algorithms that track dynamic memory allocations. That amount is difficult to estimate, but profiling seems to suggest a savings of several megabytes at least. The increased memory locality may improve the performance of the block index by some degree as well. In the future, we may refactor the data structures that contain block index objects or the allocators behind them--std::unordered_map isn't very compact. For now, this low-hanging fruit provides modest gains.
|
Some pretty clever machinery here... |
jamescowens
added a commit
that referenced
this pull request
Mar 1, 2021
Added - gui: Add RAC column to wizard summary page projects table #1951 (@cyrossignol) - rpc: clean up the superblocks function and add magnitude to getmininginfo #1966 (@jamescowens) - rpc: Add transaction size to RPC output #1971 (@cyrossignol) - voting: Add user-facing support for poll response types #1976 (@cyrossignol) - gui: Port Bitcoin Intro class (implement the ability to choose a data directory via the GUI) #1978 (@jamescowens) - gui: Port Bitcoin MacOS app nap manager #1991 (@jamescowens) - mining, rpc: Implement staking efficiency measure and improve SelectCoinsForStaking and CreateCoinStake #1992 (@jamescowens) - accrual, rpc: Implement auditsnapshotaccruals #2001 (@jamescowens) - docs: add doxygen support #2000 (@div72) - beacon: Specialized beacon storage in leveldb #2009 (@jamescowens) - rpc: Add a call to dump contracts in binary form #2011 (@div72) - rpc: Add boolean option to report active beacons only in beaconreport #2013 (@jamescowens) - consensus: Set Hilda mainnet hardfork height to 2197000 #2022 (@jamescowens) Changed - refactor: [Memory optimization] Block index duplicate PoS state #1945 (@cyrossignol) - refactor: [Memory optimization] Block index superblock and contract flags #1950 (@cyrossignol) - refactor: [Memory optimization] Remove stake modifier checksums #1954 (@cyrossignol) - refactor: [Memory optimization] Block index allocation overhead #1957 (@cyrossignol) - refactor: [Memory optimization] Remove block index subsidy fields #1960 (@cyrossignol) - refactor: [Memory optimization] Separate chain trust from the block index #1961 (@cyrossignol) - refactor: [Memory optimization] Eliminate padding between block index fields #1962 (@cyrossignol) - beacon, gui: Add check for presence of beacon private key to updateBeacon() #1968 (@jamescowens) - util: Enhance ETTS calculation #1973 (@jamescowens) - refactor: Use new clamp in util.h #1975 (@jamescowens) - gui: Redo global status for overview #1983 (@jamescowens) - util: Improvements to MilliTimer class and use in the miner and init #1987 (@jamescowens) - rpc: Move rpc files to directory #1995 (@Pythonix) - rpc: Enhance consolidateunspent and fix fee calculation #1994 (@jamescowens) - contract: Double the lookback scope of contract replay #1998 (@jamescowens) - net: Don't rely on external IP resolvers #2002 (@Tetrix42) - beacon: Change beacon map to pointers #2008 (@jamescowens) - gui: Update bitcoin_sv.ts #2014 (@sweede-se) - util: Update snapshot URLs and add accrual directory #2019 (@jamescowens) - beacon: Tweak BeaconRegistry::Revert #2020 (@jamescowens) - rpc, qt: bump fees @2023 (@div72) Removed - researcher: Remove automatic legacy beacon key import #1963 (@cyrossignol) - util: Revert "Close LevelDB after loading the block index" #1969 (@cyrossignol) - ci: Fix python symlink issue & remove travis #1990 (@div72) - ci: remove python workaround #2005 (@div72) Fixed - gui: fix mandatory/leisure detection of upgrade check #1959 (@Pythonix) - voting: Fix title in "gettransaction" RPC for legacy poll contracts @1970 (@cyrossignol) - gui: Fix missing menu items on macOS #1972 (@scribblemaniac) - rpc: Fix answer offset in "votedetails" #1974 (@cyrossignol) - voting: Implement missing try-catch in VotingVoteDialog::vote #1980 (@jamescowens) - scraper: Add check for minimum housekeeping complete in scraper #1977 (@jamescowens) - voting: Fix nonsense vote weights for legacy polls #1988 (@cyrossignol) - voting: Fix incorrect field returned in ResolveMoneySupplyForPoll() #1989 (@cyrossignol) - consensus, accrual: Fix newbie accrual #2004 (@jamescowens) - log: grammar correction #2016 (@nathanielcwm) - wallet: Correct nMinFee fee calculation in CreateTransaction #2021 (@jamescowens) - rpc, miner: Correct GetLastStake #2026 (@jamescowens) - wallet: Fix bug in CreateTransaction causing insufficient fees #2029 (@jamescowens)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This is part of a series of changes that optimize the memory usage of Gridcoin's block index. According to Valgrind's heap profiler (Massif), these changes will decrease memory usage of the application by over 500 MB after the final PR merges. Because these optimizations apply to every block, memory savings will continue to accrue as the chain grows. I'm breaking-up the commits into separate PRs because the branch has become too unwieldy to review in one submission.
This includes two optimizations that reduce the allocated memory overhead of the block index:
By allocating memory for researcher-related fields only for blocks that claim research rewards, we add 8 bytes of overhead on 64-bit systems for researcher blocks, but we save 24 bytes per non-research block. Researchers only produce about 25% of the blocks on mainnet so this change reduces memory usage by dozens of MB.
I also changed the allocation strategy for block index objects to use a pool that pre-allocates these instances in batches. The default allocator implementations likely add at least 8-16 bytes of overhead per dynamic allocation to track the regions of allocated memory on 64-bit platforms and may reserve more memory than requested. When we allocate objects in large chunks, we minimize that administrative state because the allocator tracks only the chunk instead of the memory for individual instances.
Memory allocation for the block index's
std::unordered_mapsuffers from the same problem, but changing that data structure is much more involved. We may address this in the future.