You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Downloading and processing 50 deltas that have surpassed 1MB of size takes about 5 minutes (including the download).
That gives, about 10 deltas per minute.
Each delta would have around 4000 pins. So we are processing 600 items per second, including callback notifications to the pintracker. That seems like a very low figure and we should strive to increase it. The more we do, the faster we can sync new nodes.
Is the bottleneck in the pin-tracker notifications? Or in the applying of the deltas? Or somewhere else?
The text was updated successfully, but these errors were encountered:
I noticed that this is slow too while I was using my first version of my collab cluster where I was pinning each file individually.
At least for onboarding we could fetch all deltas and patch the status until we reach the last version (at the time we started) and then notify with the calculated result?
This avoids that we're calling the pin-tracker with half-calculated results which will be overridden just seconds later (at least that's what I understand from the current approach).
We could also use this approach if we need to "catch up" more than one delta, as we would calculate all deltas, merge the results and then notify the tracker.
Downloading and processing 50 deltas that have surpassed 1MB of size takes about 5 minutes (including the download).
That gives, about 10 deltas per minute.
Each delta would have around 4000 pins. So we are processing 600 items per second, including callback notifications to the pintracker. That seems like a very low figure and we should strive to increase it. The more we do, the faster we can sync new nodes.
Is the bottleneck in the pin-tracker notifications? Or in the applying of the deltas? Or somewhere else?
The text was updated successfully, but these errors were encountered: