2010-10-15.

The earlier annoyance of very long delays (tens of seconds) in doing 
things like mkdir was apparently connected with ext3 plus large disks.
The use of ext4 has apparently helped.

But, on my systems with an NFS client and server, there's still a 
pathetic, highly embarrassing (to usual enthusiasts of the system) 
delay when doing heavy writing: things go quickly for a while, then 
at some point (presumably when some degree of memory is used and 
has to be flushed to disk? or ...?) there is a delay, not just of 
that operation by of others read or write on at least that filesystem.
When it involves NFS homedir things, this of course means practically 
everything is waiting.  It's absurd: several seconds at a time.

These might be connected:

http://www.gossamer-threads.com/lists/engine?do=post_view_flat;post=1050744;page=2;sb=post_latest_reply;so=ASC;mh=25;list=linux

http://marc.info/?l=linux-kernel&m=123732127919368&w=2

http://lkml.org/lkml/2008/10/1/405


It happens at home, with the desktop 'box' (gentoo) through Gb ethernet 
to server 'serv' (gentoo) that runs a ~2.7TB array on /home/public
and 45GB on /home, both raid 5 over 4 disks
dev/md0 /home ext4 rw,errors=remount-ro,usrjquota=.aquota.user,jqfmt=vfsv0,user_xattr,acl,data=ordered,journal_checksum,barrier=1,stripe=12,commit=4,stripe=12 0 0
/dev/md1 /home/public ext4 rw,errors=remount-ro,user_xattr,acl,data=ordered,journal_checksum,barrier=1,stripe=48,commit=4,stripe=48 0 0

It happens at work, with e.g. desktop 'gnu' (gentoo) through Gb (old 
room) or 100Mb (current room, till next weeks' upgrade) to server
'penguin' (RHEL 5).