Disk adapter failure handling by HACMP/6000

ITEM: RTA000095929



                                                                                
Could you please assist me with the HACMP/6000 capability against disk          
adapter failure?                                                                
In the following sample configuration, please teach me how you make             
HACMP detect the disk adapter(a) or cable failure and make system B             
take over for system A (please assume this configuration as mode 1).            
Or, if it's not possible by HACMP, why not?                                     
                                                                                
  -------O--O--------------------------------O--O------------- En               
         |  |                                |  |                               
      ------------      heartbeat        ------------                           
      | System A |o---------------------o| System B |                           
      |          |         9333          |          |                           
      |          |     -------------     |          |                           
      |        -----|  |  |  |  |-----|          |                          
      ------------     -------------     ------------                           
                                                                                
                                                                                
ANSWER                                                                          
HACMP/6000 is not designed to directly handle disk or disk adapter              
failures. These should be taken care of through careful hardware planning       
and setup, and using the LVM mirroring facilities of AIX. Disks (and disk       
adapters) are often viewed as more likely components to fail than CPU's,        
so an HACMP cluster should always be designed with the shared volume            
group(s) mirrored. This mirroring will be most effective if it is done          
across separate disk strings (to automatically handle the failure of a          
disk adapter) as shown in the modified diagram below:                           
                                                                                
  -------O--O--------------------------------O--O------------- En               
         |  |                                |  |                              
      ------------      heartbeat        ------------                           
      | System A |o---------------------o| System B |                           
      |          |         9333          |          |                           
      |          |     -------------     |          |                           
      |        -----|  |  |  |  |-----        |                           
      |          |     -------------     |          |                           
      |          |         |  |          |          |                           
      |          |        MIRROR         |          |                           
      |          |         |  |          |          |                           
      |          |     -------------     |          |                           
      |        -----|  |  |  |  |-----        |                           
      ------------     -------------     ------------                           
                           9333                                                 
                                                                                
If you were going to use your original configuration, you could mirror         
within your disk string, but would be exposed to a power supply failure         
on the 9333 which would cause the cluster to fail. Since each 9333              
adapter has 4 connectors, adding a second 9333 tower connected to the           
single 9333 adapters in each machine, with disks mirrored across the            
strings, would protect you from the failure of either 9333 tower. Using         
the Error Notification facility of HACMP 1.2 and beyond, you could have         
the system respond to an error log record of a 9333 adapter failure by          
causing a node failure, and moving access to the disks over to the other        
system with the good disk adapter.                                              
                                                                                
S e a r c h - k e y w o r d s:                                                  
HACMP DISK ADAPTER FAILURE MIRROR MIRRORING                                     
                                                                                
                                                                                
                                                                               


WWQA: ITEM: RTA000095929 ITEM: RTA000095929
Dated: 11/1996 Category: ITSAIHA6000
This HTML file was generated 99/06/24~12:43:32
Comments or suggestions?
Contact us