I'm going to put 4 million files into an EXT4 partition. I have about 700 files in each dir, the average file size is 38kb, total size is 169 gigabyte.

What are the best options in terms of block size, inode size and inode ratio I can choose?

Is it better to create two or more partitions, thinking at the checking time fsck could take?

  • 1
    As fsck is a concern, did you consider ZFS?
    – jlliagre
    Commented Oct 20, 2018 at 21:26
  • I thought to EXT4 because it is more known and supported I believe, I don't know the current ZFS status.
    – cdarwin
    Commented Oct 20, 2018 at 21:30
  • 1
    With using ZFS, inode size/ratio and fsck time are all irrelevant. About support, I believe only Ubuntu formally supports ZFS, although it might be installed manually on most current Linux distributions / kernels. Alternatively, there are also FreeBSD and Illumos based OSes with native ZFS support. You might also want to have a look to other file systems like XFS.
    – jlliagre
    Commented Oct 20, 2018 at 22:27
  • There's also the "use faster hardware and be done with it" approach - consider putting your data on SSD(s). Commented Oct 21, 2018 at 12:28

1 Answer 1


Handling 4M files in a single filesystem is no problem for ext4, so long as the filesystem is formatted with enough inodes. It is no problem to even have 4M files in a single directory, if the filenames are not excessively large.

There are Lustre filesystems with 1.5-2B files, and 10-12M files in a single directory (which is about the directory limit until kernel 4.recent when the "large_dir" feature was added). That means you don't need to do anything special with the directory structure to handle the files, unless you might need to store many more files in the future, or if you have a regular turnover of files, where you might want to make "age" based directories and then delete them after some time.

Reasonable formatting options would be:

mke2fs -t ext4 -i 32768 -b 4096

-i 32768 = average file size is 32KB, to ensure enough inodes
-b 4096 = blocksize, to allow large directories

The default inode size is fine, unless you store a lot of xattrs on each file. If yes (use getfattr -d -m- -ehex /path/to/existing/file to see what the average xattr size is), then use -I to increase it. The core inode size is about 180 bytes these days, and the rest is available for fast xattrs.

If you put the filesystem on an LVM/DM device, then you can also resize it online to add more space/inodes if you need more in the future. What you can't easily change is the inode ratio or inode size.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .