MCEventAccumulator : Tracking the history of Minecraft Worlds with Datomic

Sammanfattning

Minecraft is a game where “no one can tell you what you can or cannot do”. This means you can build anything, but also destroy anything. To combat destruction, server logging tools are created to roll back damage. These require a lot of storage space however, which is why this project investigates how the Datomic database performs in this context. Datomic was chosen due to its own built-in time travel system in the hopes that it would outperform other tools in terms of storage efficiency as they need to store a log of events in the database manually. Additionally, little to no research has actually been done publicly on Datomic itself, making it an interesting target for a thesis. To accomplish this, a custom logging tool was built from scratch designed specifically to work with Datomic. The tool, known as MCEventAccumulator, contains a translator to translate NBT data (Minecrafts internal data format) to the equivalent Datomic data and vice versa, as well as a basic logging and rollback system. The results indicated that MCEventAccumulator was slightly slower than Ledger (an existing logging tool for the Fabric platform) and sometimes slower than CoreProtect (the most popular logging tool for the PaperMC platform, although CoreProtect was potentially slowed down by the Cardboard compatibility layer) both in terms of overhead and rollback speed. Although it should be noted that the overhead measurement yielded an apparent negative overhead on the primary desktop despite repeated measurements. However, the storage measurement results showed that both Ledger and CoreProtect had significantly better storage efficiency. Further investigation implied that Datomic was implicitly introducing additional redundancy. While MCEventAccumulator’s implementation is partially to blame, this mistake is very easy to miss. An additional measurement was made without ownership tracking which did show an improvement, but this cannot be taken at face value as Ledger’s and CoreProtect’s databases both include this data. The final conclusion is essentially that Datomic has potential as an alternative, but further research is necessary. If anything, the main knowledge gained from this thesis is that Datomic is hard to use correctly. This thesis highlights potential pitfalls to help others avoid them.