Re: [ossig] Transfering Large Files via NFS mount

[Colin Charles <byte@aeon.com.my>]

On Wed, 2003-10-22 at 17:10, Nicholas Adrian Suppiah wrote:
>  
> If it still doesnt work, ext3 could be the problem. Try mounting with the options (-o) in the case of fstab file as follow:
>  
> 1. if you suspect that journaling is the problem
> use data=writeback instead of data=journal

really, this is probably *not* the cause. i've used nfs on ext2, ext3,
and xfs (no, no reiserfs for me). data journalling is fine.

> 2. improve the read and write rate (read being half of write). Change the values as required until things improve in speed and performance.
> use rsize=1024,wsize=2048 in fstab

this on modern machines can be *improved*.

1024/2048 are really small blocksizes. the default is in actual fact
either 2 or 4KB nowadays (long gone are the days where alan cox himself
decided that 1KB is the way to go).

i suggest rsize/wsize values to be about 8192 at the very least. please
feel free to improve in value, but too great a jump will cause a
performance hit as well. a fine balance depends on the kernel you're
running as well as the network and the hardware you have.

> or at CLI
> /sbin/elvtune -r 1024 -w 2048 /dev/sdd

erm, this only does stuff to a local disk. in kernel 2.4 (and onwards),
this basically tunes the disk i/o elevetors. it really has no effect on
nfs. if you're suggesting his local disk can't keep up, i don't think
that's a wise assumption to make.
 
> refer to http://www.redhat.com/support/wpapers/redhat/ext3/ext3.pdf

that said, i didn't read the RH whitepapers...

oh, btw, elvtune might be on RH by default, but it may not be on
mandrake. 
-- 
Colin Charles, byte@aeon.com.my
http://www.bytebot.net/


------------------------------------------------------------
To unsubscribe: send mail to ossig-request@mncc.com.my
with "unsubscribe ossig" in the body of the message