WHY CAN'T EXECUTABLES be on the shared disk?

ITEM: RTA000041834



QUESTION:                                                                       
Can you tell me why for HACMP/6000 the executables for a program                
cannot be on the shared disk? Please elaborate with examples, etc.              
The guidelines I have states that only data can be on the shared disks.         
I am sure the programs will run being on a shared disks and will                
probably have to be started manually after a takeover, but there must           
be a way to automatically start a database for example whose executables        
are on shared disks.                                                            
                                                                                
---------- ---------- ---------- --------- ---------- ----------                
A:                                                                              
Current HACMP/6000 training emphasizes a configuration prohibition on           
placing binary executables on shared media claiming that "problems              
would result" if this program was running at the time that a fail-over          
occurred.  From a conventional perspective, I can't understand why a           
read-only disk copy would be altered in any way.  The copy in memory            
would be lost, as  well as all of its pointers and file-opens, but when         
a new copy was read into a new memory on the acquiring machine, the end         
result doesn't seem any different to me than if it had been read from           
one of its own disks or the shared disks.  Could you please explain the         
rationale for this advice?                                                      
                                                                                
 Answer:                                                                        
                                                                                
 It is recommended that executable modules are not placed and run               
 from a shared disk for the following reasons:                                  
                                                                                
 A. Licensing                                                                   
      Even though the instance of an application is running on a                
      single CPU at any one time, some SW vendors take issue with the          
      potential of a single copy of the application running on multiple         
      CPUs.  Some vendors require a unique copy for each CPU that will          
      run the application, and even go as far as to license protect the         
      application by incorporating CPU specific information (Vital              
      Product Data) into the application at install time. This would            
      prevent a "fallover" CPU from restarting the application in               
      response to a failure.                                                    
                                                                                
      It is prudent to query vendors during the planning stage to               
      identify such restrictions.                                               
                                                                                
 B. Application Invocation                                                      
      Some applications (such as databases) have configuration files            
      that are tailored during installation, and stored with the                
      binaries.  These configuration files usually provide startup             
      information, such as databases to be loaded, log files to be              
      opened, etc.                                                              
                                                                                
      To put these configuration files on a shared filesystem usually           
      requires additional tailoring.  Logic needs to be added to                
      determine which system is actually invoking the application.              
      This issue becomes particularly acute in Mutual Takeover scenarios,       
      where both systems are running different instances of the same            
      application (different databases), and would be standing by for           
      one another.  Conflicts in the location and access of control files       
      can occur.                                                                
                                                                                
      Much of the tailoring can be avoided by placing slightly different        
      startup files on local filesystems on either system.  This would          
      allow context information to be static, instead of calculated each       
      time the application is invoked.                                          
                                                                                
 C. Reintegration                                                               
      When a previously failed node is ready to rejoin the cluster,             
      the other CPU must gracefully give up the disk resources it               
      had acquired at the time of fallover.  In  order to do this,              
      it must unmount the file systems and varyoff the volume groups.           
                                                                                
      If there are active binaries on the shared disk, it is necessary          
      to shutdown the processes and remove the users from the file              
      systems in order for the unmount to succeed.                              
                                                                                
      We have experienced some applications that have difficulties              
      in being shutdown gracefully.  If a binary process cannot                 
      be killed with a supplied command, the fallover server is unable         
      to unmount the the shared filesystem in preparation for the               
      varyoff of the volume group and the giveback of resources.                
                                                                                
      This will cause the reintegration process to fail and                     
      manual intervention is required.  This is an issue to be                  
      researched with the provider of the application.                          
                                                                                
                                                                                
                                                                                
---------- ---------- ---------- --------- ---------- ----------                
                                                                                
                                                                                
This item was created from library item Q658116      CRDFN                      
                                                                                
Additional search words:                                                       
CRDFN DISKS EXECUTABLES HACMP IX MAY94 OP OZNEW PROG PROGRAM                    
PROGRAMMABLE PROGRAMMER PROGRAMMING RISCOSO RISCSYSTEM SHARE                    
SOFTWARE SYS 6000                                                               
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                                
                                                                               


WWQA: ITEM: RTA000041834 ITEM: RTA000041834
Dated: 11/1996 Category: RISCOSO
This HTML file was generated 99/06/24~12:43:16
Comments or suggestions? Contact us