mlocate

I recently tried out mlocate, a new implementation of the locate utility, and I was pleasantly surprised. The 'm' in mlocate stands for "merging," which refers to updatedb's ability to reuse the existing database and avoid rereading most of the file system. This not only makes updatedb much faster, but also helps prevent system caches from getting bogged down.

One thing I appreciated about mlocate is that it aims to be completely compatible with slocate. To test its performance, I booted my computer into single-user mode and ran updatedb once, then collected data using slabtop and free. While this method may not be perfect, I believe that the resulting numbers are a good representation of real-world performance.

According to my measurements, mlocate outperformed slocate by a significant margin. On the first run, mlocate took only 1m11.65s to complete, compared to slocate's 1m32.84s. And on subsequent runs, mlocate only took 37.64s, while slocate took 0.704s. These times were accompanied by real, user, and system data reported by the time function.

In addition to the time measurements, I also collected data on the dentry and inode numbers of active objects in the dentry_cache and ext3_inode_cache (as reported by slabtop) and the cached size of disk buffers and page cache (as reported by free).

Overall, I highly recommend mlocate for anyone looking for a fast, efficient implementation of the locate utility. Just be aware that the first run will take longer as the whole file system is scanned, but subsequent runs will be much faster thanks to the reusable database.