Decoding some fields from the mon_instrum diagnostics class

ITEM: RTA000071377

  When generating a report from the ENCINA monitor's instrumentation            
data the following fields appear:                                               
                                                                                
 Sum of clock time squares:                                                     
                                                                                
 Sum of CPU time squared:                                                       
                                                                                
 The manual pages indicate that these values are as their name                  
implies, the sum of CPU and clock times squared.                                
                                                                                
  Unfortunately, my statistics knowledge is very rusty                          
and I am not sure what I am supposed to infer from these values.                
Presumably, the values reported here will depend on the application             
server being run, so how are we supposed to know what a good or bad             
value range is? Is there some guidelines that we should apply against          
these values ?                                                                  
                                                                                
Answer:                                                                         
                                                                                
These fields are in fact intended to allow the applications                     
designer/programmer a means to compute standard deviation statistics            
on the application.  In general, these data items will be useful for            
determining relative performance and providing input for system tuning          
and load balancing.                                                             
                                                                                
Unfortunately, there isn't any set of rules for determining what                
'good' or 'bad' values are.  As is often the situation, these numbers           
are highly dependant on the environment and application which they are          
measuring.  A larger than "normal" standard deviation (where "normal"           
is some average number taken over time or a comparison against a like          
server for instance) would typically indicate some sort of problem.             
This could be a problem with the RPC mechanism, a network problem               
(collisions or retransmissions) or simply not enough PAs for the                
incomming load of client requests.  Other specific items which will             
certainly effect your standard deviation would be the number of                 
clients, the number of servers, the length of time which an RPC is              
active at the server, and the scheduling order.                                 
                                                                                
S e a r c h - k e y w o r d s:                                                  
____MON_INSTRUM SQUARED AIX ENCINA MONITOR                                      
                                                                                
                                                                                
                                                                                
                                                                                
                                                                               



WWQA: ITEM: RTA000071377 ITEM: RTA000071377

Dated: 05/1995 Category: ITSAI6000EN

This HTML file was generated 99/06/24~12:43:26

Comments or suggestions?
Contact us