Description of problem:
On this machine there is running a server written in Java.

It is running stable for about one or two days and then it hangs. Our 
research revealed, that it called a native method, that tried to read 
from a file on the harddisk (it's an SCSI device) and that method 
does not return. (We observed a case where it returned with 
successful reading after about 5 minutes.)

It is everytime a different file and the file can be read without 
problems after restarting the server.

There are some strange messages in the syslog:
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Attempting to abort cmd 
f6e79600: 0x2a 0x0 0x3 0xdc 0x3 0x37 0x0 0x0 0x8 0x0
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Command already completed
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Attempting to abort cmd 
f6e79800: 0x2a 0x0 0x4 0x64 0x4 0xb7 0x0 0x0 0x8 0x0
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Command not found
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Attempting to abort cmd 
f6e79a00: 0x2a 0x0 0x4 0x68 0x1 0xc7 0x0 0x0 0x8 0x0
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Command not found
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Attempting to abort cmd 
f6e79c00: 0x2a 0x0 0x4 0x6c 0xa 0x87 0x0 0x0 0x8 0x0
Nov 17 08:54:28 Troll kernel: scsi0:0:0:0: Command not found
[snip]


Version-Release number of selected component (if applicable):
kernel-2.4.22-1.2197.nptlsmp

How reproducible:
Apears every 1 or 2 days.

Steps to Reproduce:
1. Exact triggering conditions unknown
2.
3.
  
Actual results:
Hangs for long time.

Expected results:
Reading data from Harddisk should not be blocking so long.

Additional info: