How do you solve slow file deletion on ext4?

[Roxor]

@EasyRhino

So, DisaSoft made a suggestion of just increasing the number of directory used by storj to reduce the number of files per folder…

That will probably make things slower… but it’s easy to check. They suggested having more directories (with fewer files-per-directory) but they didn’t actually test spreading the same number of files over more directories.

@EasyRhino

Do we know, with the storj nodes right now, if you increase the size of a node (say from 1TB → 10TB) does the number of directories increase, or just the number of files per directory?

A node gets larger by storing more .sj1 files in the same config/storage/blobs/[satellite]/[prefix] directories. No more layers of subdirectories.

@EasyRhino

Because if it’s files per directory then ext4 (and maybe other file systems) could indeed hit a tipping point where performance collapses when doing file operations.

Storj tailored the algo that splits all the files over those directories to not hold more than modern filesystems support. I think Pentium100 recently posted a 50TB node with no issues with numbers of files. But you’re right that as you simply have more files… all housekeeping tasks take longer. At some point (if you don’t have metadata in RAM/SSD) you’ll spend all your time running filewalkers . Another reason to run around 20TB/node these days?