ITEM: RTA000052183

A RS/6000 580H system running AIX 3.2.5 and DB2/6000 crashed last               
week with an error description of "data storage interrupt, processor"           
with a failure cause listed as "software problem".  The system dump             
did not complete.                                                               
Since this box only is only used for DB2/6000, we think that might              
have caused the crash.  The system has an average workload of 8-12              
users, and we are wondering if the box is undersized or under-                  
configured.  Could you please give me your thoughts on the following            
 1) This system has 64 MB of real memory.  Being a database server,            
    I'm sure that more memory will translate to better performance,             
    but is 64MB even enough to run DB2/6000 adequately for these users?         
 2) This system has 144MB of paging space defined.  What amount of              
    paging space is needed for DB2/6000?  Is this 144MB adequate?               
    Would running out of paging space while running DB2/6000 crash              
    the system and give the above error message?                                
---------- ---------- ---------- --------- ---------- ----------                
A: A 580H is certainly fast enough to handle DB2/6000 for the                   
   8-12 users under "normal" conditions.  I am assuming a                       
   good database design, moderate database size, well chosen                    
   indexes....  The 64MB memory should also suffice but is more                 
   problematic.  For large databases, or queries that cause                     
   table scans, or have large answer sets, you may very well                   
   need more memory.  While lack of memory can easily cause poor                
   performance, it should not cause dumps and messages of the                   
   type you are reporting.                                                      
   I suspect the database server has not had its memory management              
   buffer configuration tuned.  The major parameters that need                  
   tuning are "buffpage" and "sortheap".  There is a HONE item                  
   CVKNR that describes the procedure for tuning these.                         
   These parameters give DB2 permission to grab larger amounts                  
   of memory for holding onto data pages, and doing in-memory sorts.            
   If you are running with the system defaults you may not be                   
   efficiently using the 64MB memory that you have available.                   
   I see that you are registered for AIX Specialist phone support.              
   Please call me at 1-800-AIX-SPEC this week while you are at                 
   the console of the server, and I can talk you through tuning                 
   these parameters. (I will be out of town next week).                         
   I also want to order a PTF that will upgrade the system to                   
   the most recent maintenance level.                                           
   You also need to change the SMIT userid settings for the                     
   DB2INSTANCE userid so that it can create large CORE files.                   
   If this happens again we need a complete core dump for                       
   investigation.  The "data storage interrupt, processor"                      
   message has many causes, from bad cpu planar board, bad                      
   ethernet card, to software problems.  A valid dump will be                   
   needed if this reoccurs.  A search of reported problems does                 
   not show this to be a known problem in DB2/6000.                             
   Other points to consider:                                                   
   Was "runstats" run after creating the database to ensure the                 
   query optimizer has valid database sizing statistics.  If this               
   is not the case then every query is causing a table scan.  On                
   larger tables this could eat all your memory.                                
   144 MB paging should be plenty, but check the paging utilization             
   with the command "lsps -a". Check the  used column.                          
---------- ---------- ---------- --------- ---------- ----------                
This item was created from library item Q675812      FFDMW                      
Additional search words:                                                        

WWQA: ITEM: RTA000052183 ITEM: RTA000052183
Dated: 01/1995 Category: RISCDB2
This HTML file was generated 99/06/24~12:43:20
Comments or suggestions? Contact us