Performance benchmark results for automated tape library high retrieval rate applications. Digital check image retrievals
J. J. Gniewek, G. Davidson, B. Caldwell · 2003
Benchmark tests have been designed and conducted for the purpose of evaluating the use of automated tape libraries in on-line digital check image retrieval applications. This type of application is representative of digital library applications requiring service to multiple concurrent users with service request response time identified as a very important metric. The benchmark test design was guided by a simplified theoretical queuing performance model developed to predict average response time as a function of request rate, object size retrieved, hardware characteristics, and library configuration. An exponential interarrival time distribution was used to simulate the multiple user environment, thus corresponding to an M/E/sub k//c queue. A series of nine benchmark test runs, each consisting of nominally 1000 retrievals, was conducted for average request rates of 50, 100, 150, 200, and 250 requests per hour and for library configurations varying from two to six drives. The resulting experimental response time distributions were analyzed by fitting the data to a Shifted Erlang distribution function and also by determining the time for which the response corresponded to cumulative percentiles ranging from 10% to 95%. Good agreement was obtained between the modelled and experimentally determined average response time. For the hardware and software systems used in these tests, an average response time of approximately 30 seconds has been achieved under conditions of Poisson arrivals at average rates up to approximately 150 requests per hour. The use of timing trace data has identified the potential for further response time improvements.