The file system for avatars

Google reveals nothing, probably ready present. While the knee was born the following decision.
\r
1) In the main database (mysql) to store all the meta info + key for pictures. The key is (the name of the sqlite file + id)
\r
2) There are a bunch of small (100-200MB) files sqlite are blobs stored image.
\r
Since the write is append only and the previous doesn't change, then a daily backup will only get the last sqlite file.
\r
What do you say? nonsense?

And I honestly would advise base, recently copied its these "small image" — 600K they are there — took a day. Then we were — the two felts on, roofing felts on three gigs
Another horse well-practiced tarfs — but it does readonly

long ago the answer to this question could be reiserfs

Btrfs + storing multiple images in a single file and specifying the item in the url or somewhere else. Read about the facebook architecture and contact, there is a lot of useful information.

Think about what avatar is most likely a lot of read requests.
When you return them from the database you will rape the disk subsystem, while when you return from disk _all_ avatars (depending on the server memory) can be cached in system memory and their impact will be turbojet.

Pictures should be stored on the hard disk and no data base will not fit here because it creates unnecessary load on the server. And what about how to store it, I would recommend to split into subfolders by id-photo like this: if pictures < 100k, that's enough /ff/32123e21.jpg. If > 100k && < 2m, then /ff/ff/321312312.jpg well, and c /ff/ff/ff/321312312.jpg in space...

And, depending on the needs, you can use a database like Derby — it is in itself always consists of tens of files, regardless of the number of tables in it