Lecture 9
Bounds on Throughput, Performance Engineering Methodology

• We reviewed the asymptotic bounds on throughput under light and heavy load conditions:
  • Heavy load (the utilization of a resource cannot exceed 100%):
    

    
    		X_0 = U_i / D_i
    
    	            <=  1 / D_i
    
    	

  • Light load (considers # concurrent jobs and time for service, but ignores waiting time)

    
    		N =  R * X_0  >=   Sum, i=1 to k (D_i) * X_0
    
    		  =>  X_0 <=  N /  Sum, i=1 to k (D_i) 
    
    	

• We looked at an example using excel: perfbounds_09_09_08.xls
• We began to look at performance engineering methodology in the context of an example: a call center. Given the example, we answered some questions about the center:
  • If 200 customer service reps work for the company, and 80% of them are working during the peak hour, and the average think time is 30 sec, what should the system throughput of the application server be to support sub-second response time to the customer service reps? ... solved using the interactive response time law ...
  • What should the capacity of the DB server be to support this? ... solved using the Forced Flow Law ...