Description of problem:
The system becomes unresponsive during periods of heavy i/o.

This bug began life as service request 345034.  I tried using and
ext2 filesystem in place of ext3 and (not unexpectedly) halved
i/o's but no real change in response time to commands from the
shell prompt, telnets into the system, etc.  Am including the
original description of the service request:

Our application does large amounts of i/o (writes out 10GB to a file, 
reads 
similar amounts from a local file).  During these periods, any other 
activity 
on the system becomes impossible (such as an ls, a remote login, 
etc.).  I 
was wondering if you have any tuning recommendations that you could 
forward. 
I've checked through the support database, but didn't see anything.  
I've 
been changing system parameters such as /proc/sys/vm/bdflush, using 
elvtune, 
along with old unix tricks like noatime on file systems, but haven't 
really 
gotten the right combination.


Version-Release number of selected component (if applicable):
e49enterprise

How reproducible:
Very

Steps to Reproduce:
1.Run something that generates a heavy write load (such as the 
attached program).
2.Try to do an ls from a command prompt; try telneting to the 
machine
3.
  
Actual results:
Response time to commands on the order of minutes

Expected results:
Response time to commands in the seconds range.

Additional info:
System is a Compaq Proliant DL380G3 with dual 3.06GHz Xeons with
9GB of memory writing to locally mounted SCSI drives.