We've found that lots (tens of thousands) of large blobs (tens of kbytes)
that change frequently (daily to weekly basis) are _evil_, at least under
3.22.x. You tend to get a fair amount of fragmentation, and you end up very
bound to disk i/o, especially seek times. 3.23 is supposed to handle
fragmentation better, though I've not done extensive testing with it.
Keep in mind that most filesystems search their directories in a linear
fashion, which in various cases can give you O(N^2) access times. Having
more than a couple hundred files in a directory can easily turn into a
bottleneck. Squid and Apache and other programs tend to address this by
using a multi-level directory structure to distribute filesystem directory
searching into smaller chunks. O(N^2+N^2)<<<O((N+N)^2)!
----- Original Message -----
From: Travis <travis@stripped>
To: mysql <mysql@stripped>
Sent: Monday, July 26, 1999 4:04 PM
Subject: images/ video
> I am curious, which is better, to store binary media in blobs in a
> table, or to use a subdirectory with a file name that corresponds to the
> primary key of the table, and just read them from the file system. Or is
> there another option that is more favorable. Any pointers or suggestions
> will be much appreciated.