A while back when I was doing more file system stuff, I looked hard at this
area and gathered up stats from a bunch of Sun's on Sun's mountain view
campus.
This data is from '93 so I'm not positive that it is still accurate but
I'll bet long odds things haven't changed that much.
The conclusion that I came to as a result of looking at this (and a lot
more) data was that the wrong thing to do is worry about single files.
What you want to do is to figure out how to put mutiple files in one
"file" - essentially make tar be a part of the file system where you put
related files in the same "tarball". Then when you access any file -
metadata or data - then you schlep in the entire tarball and explode it
in memory. Food for thought.
Here's the data, this is distribution of files as a function of size.
It's very important to note that this was //**//NOT//**// an analysis
of on disk data sizes - I was looking at the set of files which were
actually in memory.
Size Count Log2 histogram
1 ( 48) ######
2 ( 294) #########
4 ( 430) #########
8 ( 105) #######
16 ( 922) ##########
32 ( 539) ##########
64 ( 792) ##########
128 (1454) ###########
256 (1637) ###########
512 (2024) ###########
1K (3421) ############
2K (2978) ############
4K (3268) ############
8K (3298) ############
16K (3908) ############
32K (6824) #############
64K (2934) ############
128K (2347) ############
256K (1186) ###########
512K ( 950) ##########
1M ( 850) ##########
2M ( 617) ##########
4M ( 155) ########
8M ( 59) ######
16M ( 65) #######
32M ( 141) ########
64M ( 5) ###
128M ( 5) ###
256M ( 3) ##
41259
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/