Re: double read confusion

V Ganesh (ganesh@vxindia.veritas.com)
Mon, 19 Apr 1999 01:46:41 +0530 (IST)

> From: "Peter J. Braam" <braam@cs.cmu.edu>
> Assume that next we want to write in block 1, then ext2_getblk, calls
> inode_getblk, which calls getblk, and, unless I'm mistaken, we go back
> to the _disk_ to read the block.
>
> My confusion is about the following: why can't we share page cache
> data with the buffer cache, eliminating the extra i/o?

we can. I tried this a month back, patched ext2_file_write to use
the page cache if possible rather than do the I/O. also put in a /proc
hook which showed how many I/Os we save. the result was a pathetic _32_
reads saved in a whole day of typical desktop activity (including kernel
compiles). you can of course write a program to read a block, write a byte,
read a block, write a byte etc., but such behaviour is quite rare in typical
unix environments.
anyway 2.3 shouldn't have this problem once the whole kludgy read-through-
pagecache-write-through-buffer-cache routine is replaced with
genetic_writepage and co.

ganesh

-
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/