Re: journalling options [WAS: optimizing ext3...]
On Mon, May 09, 2005 at 01:59:42PM +0100, michael wrote:
> I've been trying to get somewhere with the below (seems easier to top
> post this time - apols!) and have come across the "data=journal" mode
> for ext3.
Journaling data instead of just metadata will cause higher disk I/O, not
less.
> Is it possible to change this on a live system
Sure: mount -o remount,data=journal <device>
> Any other thoughts on why my system seems to creep to a crawl whenever
> there's IO (eg compressing a file, creating a tar, running progs with
> big (a couple Gb) files)?
Swap usage (too much/too little), memory usage, slow spindles, memory
leaks, bus contention, atime updates, NFS mounts...the list goes on.
You need to look at something a little more granular, such as sar or
iostat, to see what all the disk I/O is all about. But if you've got a
single IDE spindle at 5400 RPM, for example, what are your realistic
expectations of performance of multi-GB file manipulations?
You might also want to look at profiling your Fortran code. Sometimes,
bad I/O performance can be traced to inefficient read/write handling in
a program. YMMV.
--
Re-Interpreting Historic Miracles with SED #141: %s/water/wine/g
Reply to: