Adding a mutex lock to set_range function #4207
Closed
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
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
|
This pull request was exported from Phabricator. Differential Revision: D75555658 |
Raahul46
force-pushed
the
export-D75555658
branch
2 times, most recently
from
c967848 to
cdc6c13
Compare
|
This pull request was exported from Phabricator. Differential Revision: D75555658 |
Raahul46
force-pushed
the
export-D75555658
branch
from
cdc6c13 to
2071f80
Compare
Summary: Pull Request resolved: pytorch#4207 X-link: facebookresearch/FBGEMM#1281 **Context:** While we expose KVTensor to external surfaces (i.e., checkpointing), they have the freedom to leverage the KVTensor functions in a concurrent fashion. For example, https://www.internalfb.com/code/fbsource/[5b7b1eef7d69]/fbcode/aiplatform/modelstore/checkpointing/pyper/TensorLoaderCallback.h?lines=85-86 This function here calls set_range to the same KVTensor multiple times because we divide a huge chunk of data into smaller chunks and try to write it in a concurrent fashion. This is a bad practice because in SSD I/O, We also use multi threading to write data in KVTensor. Currently, we use 32 threads (each thread per shard) to write data. Due to this, when we call set_range multiple times, this can lead to thread contention and increase in synchronization overhead **In this Diff:** We introduce a mutex lock on the set_range function, due to this every transaction is locked during execution and the multiple calls are processed serially leading to more efficient use of the threads Differential Revision: D75555658
|
This pull request was exported from Phabricator. Differential Revision: D75555658 |
Raahul46
force-pushed
the
export-D75555658
branch
from
2071f80 to
98cc2b6
Compare
|
This pull request has been merged in c845cc9. |
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.
Summary:
Context:
While we expose KVTensor to external surfaces (i.e., checkpointing), they have the freedom to leverage the KVTensor functions in a concurrent fashion.
For example,
https://www.internalfb.com/code/fbsource/[5b7b1eef7d69]/fbcode/aiplatform/modelstore/checkpointing/pyper/TensorLoaderCallback.h?lines=85-86
This function here calls set_range to the same KVTensor multiple times because we divide a huge chunk of data into smaller chunks and try to write it in a concurrent fashion. This is a bad practice because in SSD I/O, We also use multi threading to write data in KVTensor.
Currently, we use 32 threads (each thread per shard) to write data. Due to this, when we call set_range multiple times, this can lead to thread contention and increase in synchronization overhead
In this Diff:
We introduce a mutex lock on the set_range function, due to this every transaction is locked during execution and the multiple calls are processed serially leading to more efficient use of the threads
Differential Revision: D75555658