The talk is about the work done in making LibreOffice faster by reducing allocations and converting usage of the internal OSL mutex to the standard mutex. The SVL shared string pool was identified as a bottleneck, and a concurrent lock-free hash map was implemented, but it turned out that the hotspot was in the uppercase conversion, which was later improved by another developer, rendering the hash map useless. Red line processing in writer was identified as slow, but the data structures are inherently complicated, making it difficult to speed up. The speaker considers failures as learnings and wins in their quest to make LibreOffice faster.