chore(model): improve block pull scheduler - #10678
Draft
calmh wants to merge 2 commits into
Draft
Conversation
Copy link
Copy Markdown
Member
Sorry, something went wrong.
The current scheduler is based only on the number of outstanding blocks towards each device. This will, in the long run and under most circumstances, result in using maximum throughput from each device. However, it means that when we start pulling (in each iteration) we'll distribute blocks equally regardless of device speed, and will keep requesting blocks from slower devices when we don't have to. This change improves the selection: - remember the last ten block download times and calculate average download rate by device - track the amount of outstanding request bytes per device - for a new block request, select the device with the shortest queue (in seconds) Signed-off-by: Jakob Borg <jakob@kastelo.net>
Copy link
Copy Markdown
Contributor
|
Cool :-) Any chance the queue length and/or observed rates could be exposed somehow through app.Internals? (Or more generically, a way to use this logic when invoking RequestGlobal)? As I described here there is similar logic in Synctrain, albeit with different goals (low latency and mostly sequential fetches, needs to be robust when connections are not). Tangentially, it would be nice if RequestGlobal could be told to use a wholly separate connection (for existing connections the queue may be long, while for Synctrain on-demand files should ideally be prioritized). |
Sorry, something went wrong.
Proof-of-Ineffective-Input
pushed a commit
to mindires/syncdesic
that referenced
this pull request
Jul 24, 2026
…upstream research - replace simple RTT selection with Adam momentum (beta1=0.7, beta2=0.9) for connection quality scoring (smoothed RTT + variance decay) - add section 4: upstream PR research (syncthing#10678, syncthing#8918, syncthing#10516, sushitrain) - add Adam paper reference to section 9 - update expected effects and implementation steps accordingly
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
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.
The current scheduler is based only on the number of outstanding blocks towards each device. This will, in the long run and under most circumstances, result in using maximum throughput from each device. However, it means that when we start pulling (in each iteration) we'll distribute blocks equally regardless of device speed, and will keep requesting blocks from slower devices when we don't have to. E.g., if we need four blocks and have two devices, we will almost always schedule two blocks for each device, even if one device is 100 times faster to respond.
This change improves the selection: