In the last post I described my weekend project of using advice from Tiger Style to optimize the write path of Seb. Here, we found that data serialization and memory allocations were big contributors to the application being slower than it could be, and profiling helped us identify places on the write path where batching and buffer reuse could greatly improve the throughput. With a few small changes, we doubled the number of records that Seb can write to disk per second! In this post we’re ...
