Improve dispatch queue usage #69
Open
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
|
Thanks, do you have a benchmark result? |
I used both configurations when collecting data for #58 and I don't remember any difference. I can do another run on macOS 15 and M2 Max. |
|
Rebased, I'll do quick test run to validate the lack of performance improvement and update the commit message. |
The way we used dispatch queue was not optimal, leading to creation of too many threads, which can cause unneeded context switching, and lower performance. We use now 2 queues: - vms_queue - concurrent queue use to run accepted connections threads. For this queue we must use concurrent queue since we want to process requests from VMs in parallel. - host_queue - serial queue for processing events from vmnet. This queue is used for the start and stop callback and for the events callback. With this change we use consistent number of threads: main thread, 1 vmnet thread, 1 thread per connected vm. Testing show no change in performance. | hardware | os | test | before (Gb/s) | after (Gb/s) | |----------|--------|---------------|---------------|--------------| | M1 Pro | 14.7.1 | host-to-vm | 3.252 | 3.219 | | M1 Pro | 14.7.1 | host-to-vm-2 | 1.902 | 1.880 | | M1 Pro | 14.7.1 | vm-to-vm | 1.493 | 1.507 | | M2 Max | 15.1 | host-to-vm | 2.335 | 2.359 | | M2 Max | 15.1 | host-to-vm-2 | 1.798 | 1.802 | | M2 Max | 15.1 | vm-to-vm | 1.275 | 1.266 | Results are average send throughput of 10 runs. Using performance tests from lima-vm#66. We have a performance regression on macOS 15.1. With 14.7 on same M2 Max machine it was up ot 1.2 times faster. With 15.1 it is up to 1.4 times slower. Signed-off-by: Nir Soffer <[email protected]>
|
@AkihiroSuda added test results, confirming that there is no improvement, and also showing that we have a performance regression in macOS 15.1 (not related to this change). |
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.
The way we used dispatch queue was not optimal, leading to creation of
too many threads, which can cause unneeded context switching, and lower
performance.
We use now 2 queues:
vms_queue - concurrent queue use to run accepted connections
threads. For this queue we must use concurrent queue since we want to
process requests from VMs in parallel.
host_queue - serial queue for processing events from vmnet. This queue
is used for the start and stop callback and for the events callback.
With this change we use consistent number of threads: main thread, 1
vmnet thread, 1 thread per connected vm.
Testing show no change in performance.
Results are average send throughput of 10 runs. Using performance tests
from #66.
We have a performance regression on macOS 15.1. With 14.7 on same M2 Max
machine it was up ot 1.2 times faster. With 15.1 it is up to 1.4 times
slower.