[Disco] Reduce Process/ThreadSession message queue reads and writes #16817
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 PR reduces the number of reads and writes for the message queue of ProcessSession and ThreadSession in Disco by caching all the data to read/write. The message queue in ThreadSession prior to this PR grabs the mutex for multiple times for a batch of data to read/write. This PR enables to read/write data from/to a local buffer first, and then read/write from/to the critical region together. This reduces the number of grabbing mutex to once. The message queue in ProcessSession prior to this PR reads/writes the inter-process pipe for multiple times for a batch of data. This PR uses a local buffer to cache all the data first, and then issues a single read/write from/to the pipe, and effectively reduces the number of reads/writes to the pipe, which may causes extra system overhead.
tqchen
approved these changes
Mar 29, 2024
Contributor
Author
|
@tvm-bot rerun |
thaisacs
pushed a commit
to thaisacs/tvm
that referenced
this pull request
Apr 3, 2024
…pache#16817) This PR reduces the number of reads and writes for the message queue of ProcessSession and ThreadSession in Disco by caching all the data to read/write. The message queue in ThreadSession prior to this PR grabs the mutex for multiple times for a batch of data to read/write. This PR enables to read/write data from/to a local buffer first, and then read/write from/to the critical region together. This reduces the number of grabbing mutex to once. The message queue in ProcessSession prior to this PR reads/writes the inter-process pipe for multiple times for a batch of data. This PR uses a local buffer to cache all the data first, and then issues a single read/write from/to the pipe, and effectively reduces the number of reads/writes to the pipe, which may causes extra system overhead.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
This PR reduces the number of reads and writes for the message queue of ProcessSession and ThreadSession in Disco by caching all the data to read/write.
The message queue in ThreadSession prior to this PR grabs the mutex for multiple times for a batch of data to read/write. This PR enables to read/write data from/to a local buffer first, and then read/write from/to the critical region together. This reduces the number of grabbing mutex to once.
The message queue in ProcessSession prior to this PR reads/writes the inter-process pipe for multiple times for a batch of data. This PR uses a local buffer to cache all the data first, and then issues a single read/write from/to the pipe, and effectively reduces the number of reads/writes to the pipe, which may causes extra system overhead.