This post is to announce OctoberPeaks 1.11 (based on OctoberProject 1.12).

No video for this release since there isn't anything visually obvious to show. All the changes were under the hood (mostly on the server side but some on the client side).

The most concrete change in this release is that the on-disk storage is now very different (directories of cuboid cluster files instead of a single directory containing per-cuboid files) and extraneous write-back is being avoided. This should avoid possible performance problems when the directory gets very large. When loading an old world, a conversion will be triggered.

Additionally, the default thread count for the core tick logic is determined based on how many the system can run, instead of a hard-coded default. This may improve performance for some users, but mostly for servers. Let me know if this defaulting is incorrect in any configurations.

Beyond these, the performance improvements are related to how internal components are designed or implemented, so it is hard to point to specific things which can be seen from the outside.

It is worth noting that this release was able to support stress-testing of 40 headless clients on a single server instance, on my dev machine, with hardly any (single-digit percentages, if that) dropped ticks or over-time ticks once reaching stead-state. Note that these clients are configured to each walk forward, placing blocks behind them, so they do cause world generation and updates.

While performance improvement is never "done", this release was declared since so many changes and improvements had been made since the last release that it seemed like a good point to use as the basis for new content in the next release.

The next release (1.12) is called "The Motion Update" and is planned to include blocks which can push/pull other blocks and also providing sub-block world collision for entities (allows for slabs, stairs, etc).

Let me know what you think,
Jeff.