Unfortunately the replica version that was just elected did not offer the improvements that we hoped for yet. The main issue seems to be that the changes to the scheduler indeed make it more “fair” in the sense that it cycles through the canisters much more quickly. However, this quicker cycling through canisters leads to significant memory contention on the replicas, making the actual execution more than an order of magnitude slower.
The team has ideas to address this issue and is treating this as their top priority, but unfortunately it still means that we’ll have to wait a bit longer to see the situation on busy subnets improve.
Great effort all at DFINITY who have been working on this. It’s a shame this hasn’t panned out quite as expected yet. Any ideas why the actual effect of this change wasn’t picked up during testing?