On Thu, 2002-02-28 at 18:44, Andreas Dilger wrote: > On Feb 28, 2002 13:35 +0100, Diego SANTA CRUZ wrote: > > However, I have observed that whenever something is doing a lot of > > writing to an ext3 filesystem other processes that involves I/O on the > > same disk are blocked for long times (often in the 10-30 secs range). > > For example, if I cp a 1.5 Gb file from one ext3 partition to another > > ext3 partition and I try to do ls on some dir, the ls usually blocks > > between 10 and 30 secs before giving a response (the CPU usage is no > > more than 10% during this). However if the destination partition for the > > cp is mounted as ext2, the ls command responds almost immediately. > > I don't think this is so much an ext3 problem as an elevator starvation > problem. You should try using Andrew's elevator starvation fix (sorry, > don't know URL), which allows read requests to be put earlier in the > queue so you don't get such long waits. > Thanks for the tip, I actually suspected something like this. I suppose it shows more with ext3 since it probably pushes the I/O system a little harder than ext2. Is the patch you mention the one at http://www.uwsg.iu.edu/hypermail/linux/kernel/0201.0/1654.html ? How stable is it, there was not much debate on l-k about it. Is it a good sign? Is that merged in any of the latest kernels? Thanks for the help, Diego > Cheers, Andreas > -- > Andreas Dilger > http://sourceforge.net/projects/ext2resize/ > http://www-mddsp.enel.ucalgary.ca/People/adilger/ > > > > _______________________________________________ > > Ext3-users@redhat.com > https://listman.redhat.com/mailman/listinfo/ext3-users -- ------------------------------------------------------- Diego Santa Cruz PhD. student Publications available at http://ltswww.epfl.ch/~dsanta Signal Processing Laboratory (LTS) Swiss Federal Institute of Technology (EPFL) EPFL - DE - LTS, CH-1015 Lausanne, Switzerland E-mail: Diego.SantaCruz@epfl.ch Phone: +41 - 21 - 693 26 57 Fax: +41 - 21 - 693 76 00 -------------------------------------------------------
