We are running 1.5S from around Feb 24/25 with Chuck's balancing patches.  
While performance is generally good we see some cases involving reading from 
NFS where performance is dismally bad. Not sure if this is an NFS vnlocking 
problem, UBC or what.

An example:

Client and server machines are both i386's running the same 1.5S, both with 
100MB switched network connections. Server has softdeps enabled for the 
filesystem.

Netscape Messenger on the client updating the summary file for a 35MB Inbox that
lives on the server took 15 minutes!  From ktraceing it while it was doing 
this all the netscape seemed to be doing was reading the Inbox file and 
updating its little indicators saying how far through it was.

Shortly after it had finished, a cat of the Inbox completed in much more 
reasonable time:

city-art% time cat ~/nsmail/Inbox > /dev/null
0.000u 0.075s 0:00.13 53.8%     0+0k 0+0io 0pf+0w

However if I try to do the same cat in the middle of netscape reading the file 
it also slows down (I killed it after it hadn't completed in 3 minutes).


Any ps of netscape I do while its behaving like this shows it waiting on 
"biowait" except when I also had the cat running when the cat was in "biowait" 
and the netscape was in "vnlock" any time I looked.


We have also seen similar behaviour from MH scanning folders and from other 
apps.

Any suggestions?  Should I send-pr this?

cheers
mark