Kbase P118356: Auto Trimming of AppServers causes operating system run queue to increase dramatically ( from averag
| Autor |
Progress Software Corporation - Progress |
| Acesso |
Público |
| Publicação |
07/09/2007 |
|
Status: Unverified
FACT(s) (Environment):
IBM AIX
SYMPTOM(s):
Auto Trimming of AppServers causes operating system run queue to increase dramatically ( from average of 2 to over 100 ) for a small period of time.
Disconnection of promon from database causes operating system run queue to increase dramatically ( from average of 2 to over 100 ) for a small period of time.
Database processes are connected to is using large amount of RAM.
32-bit Version of Progress / OpenEdge
CAUSE:
Due to the architecture of the AIX operating system the only way to use large numbers of shared memory segments (greater than 11) is to set a variable EXTSHM. The use of the method to make the operating system allow more than 11 shared memory segments to be used causes some additional processing requirements by the OS for memory allocation and deallocation operations.
IBM was actively involved with investigating what operations were causing the increased length of time to disconnect and end the Progress processes and they have determined that nothing is out of the ordinary when taking into consideration that the EXTSHM is active. About 4k to 7k extra memory related (specifically Virtual Memory Manager) calls were made by the operating system to deallocate memory and decommission the process.
FIX:
The suggested course of action would be to upgrade to the 64-bit version of Progress / OpenEdge where these operating system limitations no longer require the use of the EXTSHM environment variable to allow greater than 11 shared memory segments.
A second less preferred option would be to decrease the amount of memory used by the database(s) to less than what would be required to use 11 segments of shared memory (approximately 1.4 gig of RAM).
Since a process connecting to multiple databases in shared memory will need access to all the shared memory segments this might also mean a process connected to multiple databases will still encounter longer shutdown times and might trigger the OS's normal behavior to dramatically increase the run queue if connected to more than 11 segments of shared memory.