Best practices for having bunch of small files on vxfs

[Viktor Balogh]

Hi,

Could you point me some best practices for having some million small files (around ~50k each) on vxfs? The fragmentation is extensive so I'm thinking about having a cronjob with fsadm, but that's all I've found so far. I think I need to create a separate filesystem for those data. I'm speaking about Oracle audit logs, so it's mostly write-once read-never. ;) Not sure why they chose FS-side logging, don't ask me about that.

Regards,

Viktor

[Stijn Adriaensens]

We have a directory that contains 1.000.000 files somewhere, but experience a lot of performance issues when accessing it. A simple ls runs for 15 minutes. 

The maximum number of inodes is 1 billion, so it's possible, but I read somewhere that it's best not to go above 100.000 files in one directory. 

I'd advise to create a script to cleanup the directory and move older files to separate file systems, or directories.  

[Dennis Handly]

You can also tar up many small files into one bigger archive.  And gzip if you want to save space.

[Bill Hassell]

Once you have a flat directory with 100's of thousands of files (or millions), you have a major performance issue in many, many areas. A simple ls is out of the question...you have to deal with the files in a more sophisticated way. The filenames are very likely a predictable pattern (sequential numbers, sequential timestamps, whatever). If you need to move (or remove) a roup of files, start with the most restrrictive pattern you have, perhaps an exact match for all but the last 2 characters: ls abc123456?? If you have knowledge of the quantity versus filenames, use that to pick groups to move/remove.

Defrag will be very disk intensive for a long time, probably cancelling any benefit after the defrag.

But the real fix is: DON'T allow this type of design. It will always be a major problem. Create a directory structure, probably based on dates. Then script a procedure to a few hundred files to the directories. Once the directory structure has been populated, the active set of files should be a few hundred, perhaps a couple of thousand. Then create a cron script to move and cleanup the files every day.