remove freelist and buffer_strategy_lock #10
Draft
Conversation
This file contains hidden or 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
gburd
force-pushed
the
rm-freelist
branch
20 times, most recently
from
3c01069 to
52192b5
Compare
gburd
force-pushed
the
rm-freelist
branch
2 times, most recently
from
ca79be4 to
06ffe4d
Compare
Minor edits to comments only.
This set of changes removes the list of available buffers and instead simply uses the clock-sweep algorithm to find and return an available buffer. While on the surface this appears to be removing an optimization it is in fact eliminating code that induces overhead in the form of synchronization that is problemmatic for multi-core systems. This also removes the have_free_buffer() function and simply caps the pg_autoprewarm process to at most NBuffers.
Combine nextVictimBuffer and completePasses into a single uint64 called clockSweepCounter so as to eliminate the need for the buffer_strategy_lock that synchronizes their progression. Increment this counter atomically by 1 at each tick. The hand's location is the counter modulo NBuffers, the number of complete passes is the counter divided by NBuffers. Overflow of the clockSweepCounter would require ~10 years of continuous operation at ~59 billion ticks per-second, so unlikely that we consider this to be impossible.
Improve the performance of the buffer manager by replacing the modulo and division operations with a technique described in the paper "Division by Invariant Integers using Multiplication" [1]. Our implementation is inspired by the MIT Licensed "fastdiv" [2]. This algorithm provides accurate division and modulo in constant time that is pipeline and ALU friendly and estimated to take about ~12-18 cycles (vs 26-90 for hardware division). Because our divisor (NBuffers) is fixed at startup so we need only calculate the constant used by it once. [1] https://gmplib.org/~tege/divcnst-pldi94.pdf [2] https://github.com/jmtilli/fastdiv
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.
https://www.postgresql.org/message-id/flat/E2D6FCDC-BE98-4F95-B45E-699C3E17BA10%40burd.me
https://commitfest.postgresql.org/patch/5928/