Implement MultiThreadedCache.jl #1
Merged
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
This cache has no locks on access, and only has contention on a cache miss. It only ever holds a shared lock for a constant-time duration, never while executing user code. A Task requesting a key that is already being computed on another Task will block while that computation is being performed. By taking advantage of the append-only properties of a cache, the cache can be duplicated per-Thread, to avoid locking in the common case.
kpamnany
reviewed
Feb 14, 2022
tveldhui
reviewed
Feb 14, 2022
Add constructor that provides pre-computed values to the base_cache
Rethrow the exception onto the Future for all blocked tasks, delete the future, and rethrow on the current task. The cache remains usable afterwards and nothing is recorded for the key with the error.
The results are a bit lukewarm (glad we measured it), but look alright. Here's a few runs, each with different JULIA_NUM_THREADS: ```julia ┌ Info: benchmark results │ Threads.nthreads() = 1 │ time_serial = 0.013336288 │ time_parallel = 0.09071632 └ time_baseline = 0.115363526 ``` ```julia ┌ Info: benchmark results │ Threads.nthreads() = 2 │ time_serial = 0.011262138 │ time_parallel = 0.097021203 └ time_baseline = 0.139031655 ``` ```julia ┌ Info: benchmark results │ Threads.nthreads() = 20 │ time_serial = 0.011997677 │ time_parallel = 0.658225544 └ time_baseline = 1.032283809 ``` ```julia ┌ Info: benchmark results │ Threads.nthreads() = 100 │ time_serial = 0.013902211 │ time_parallel = 1.999772731 └ time_baseline = 9.314424419 ``` So it definitely does not scale as well as a single-threaded codebase would have scaled. But it also definitely scales better than the baseline, which is a Mutex around a Dict. Maybe there will be some places to improve our contention in the future :)
Gracefully handle exceptions thrown during `get!()` functions
|
Also i'm open to other package naming suggestions if anyone has them! :) |
vilterp
approved these changes
Feb 15, 2022
Add benchmark test measuring parallel scaling.
Fix lazy construction of Dicts, per guidance from Julia Base
…cache across the whole execution
...... In retrospect, this does make this whole structure start to look a lot like a Dict + a Read/Write lock, and I wonder how their performance would compare......
Concurrency safety fixes
|
After all the latest changes, especially #6 for the concurrency fixes, I think this package is good to go. It still scales a good bit better than a Dict + Mutex, so I think there's still value in it! If we want to give another approach a shot in the future, like a concurrent hash table, i'm super supportive. But hopefully this is useful in the interim. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
Implements a MultiThreadedCache{K,V}, and adds stress test.
This cache has no locks on access, and only has contention on a cache
miss. It only ever holds a shared lock for a constant-time duration,
never while executing user code. A Task requesting a key that is
already being computed on another Task will block while that computation
is being performed. By taking advantage of the append-only properties of
a cache, the cache can be duplicated per-Thread, to avoid locking in the
common case.
This PR adds: