@graf @NonPlayableClown @Tony @dj @koropokkur @verita84 Like, right now, with three machines running workers, it's averaging about 34k instances per hour, about 41% of all instances running behind schedule. It stayed at 0.2% behind schedule when it was at 75k instances per hour.

Starting up a couple of workers on a VPS or two with fat enough pipes would likely get it up past 75k; a really beefy one might make it easier to actually run it from there. On the other hand, since I moved all the data storage to Revolver, I don't know how the throughput would be affected. I think it improved it but it's not exactly an apples-to-apples comparison. I do know that it eliminated about 100GB of data from Postgres (and the block-level de-duping helped significantly).