Issue to read millions of small files form one directory

pan64 · 11-25-2022, 12:31 AM

for example ls by default sorts the filenames, so it must read all the entries of the directory before showing the content. ls -U may help on this. But probably better to use another tool to read a dir file by file.
Image processing must take much longer then fetching a filename (in general).
Without knowing actually what's going on extremely hard to catch the bottleneck. Probably it is in line 43.

purveshp · 11-25-2022, 03:58 AM

Thanks.

Any distributed filesystem or filesystem like btrfs would help?

pan64 · 11-25-2022, 04:05 AM

Quote:

Originally Posted by purveshp

Thanks.

Any distributed filesystem or filesystem like btrfs would help?

without details hard to say anything. Try it. From my side using a distributed filesystem is not the right direction, but this is the correct answer:
https://www.reddit.com/r/Jokes/comme...et_a_dinosaur/

elgrandeperro · 11-25-2022, 10:37 AM

The stat of the directory is what is taking time, not the size of files. That is why hashing the path to reduce directory size is done. I've converted a filesystem to a database lookup to fix the problem also, databases are optimized for lookups. I'd say it will never work as designed.

TB0ne · 11-25-2022, 10:50 AM

Quote:

Originally Posted by purveshp

Thanks.
Any distributed filesystem or filesystem like btrfs would help?

Again, as you've been told several times, you haven't given us any real details about what you're trying to do, how, or describe your actual goal. Saying "read small image files" tells us nothing. You hint at multiple clients (running what? Attached how), over a network (LAN? WAN? VPN? what?) to do 'manipulation' (what kind? Using what program?).

Without details and a clear question, there's nothing we can tell you that would be meaningful.

wpeckham · 11-25-2022, 11:29 AM

Quote:

Originally Posted by purveshp

Thanks.

Any distributed filesystem or filesystem like btrfs would help?

You have given us inadquate data to make any other than general comments. We need far more detail, and better detail, about your hardware and network situation, your processes, your data, files, and structures, and far more data about the processes that create or deliver these objects and about how and why you intend to process these objects. You have us shooting at goldfish, in a barrel, without lights, in the dark. Even if we his something, we will never know.

Can you give us enough information to actually work on?