tailieunhanh - THE FRACTAL STRUCTURE OF DATA REFERENCE- P25

THE FRACTAL STRUCTURE OF DATA REFERENCE- P25:For purposes of understanding its performance, a computer system is traditionally viewed as a processor coupled to one or more disk storage devices, and driven by externally generated requests (typically called transactions). Over the past several decades, very powerful techniques have become available to the performance analyst attempting to understand, at a high level, the operational behavior of such systems. | 110 THE FRACTAL STRUCTURE OF DATA REFERENCE 2. A CASE STUDY This section improves upon the analysis just presented by taking into account a more complete picture of both costs and recall delays at a specific installation. The case study presented below was performed by capturing the smf records related to storage management so as to simulate alternative storage management policies against the captured data. The installation of the case study was a moderate-sized os 390 installation with a mix of on-line cics ims and DB2 data base activity plus a small amount of tso storage. Essentially all user and database storage was SMS-managed and was contained in a management class called standard. At the time ofthe study policies in the standard pool called for migration off of level 0 storage after 15 days and migration off of level 1 storage after an additional 9 days. The smf data used in the study covered a period of 33 days. One immediate purpose of reassessing the hierarchical storage management policies at this installation was to examine a planned installation oftape robotics. The case study involved the following steps 1. Capture the daily smf 14 15 17 64 65 and other miscellaneous record types associated with storage management. 2. Extract the key smf data and accumulate at least 30 day s worth. 3. For each combination of level 0 and level 1 migration ages up to a level 1 migration age of30 days simulate the resulting migrations recalls storage requirements and costs. 4. input the simulation results into a software package capable of contour plotting. 5. Use graphical techniques as described below to perform a constrained optimization based on the costs and recall rates associated with each combination oflevel 0 and level 1 migration ages. Steps 1-3 were performed using the sms Optimizer software package 42 . The cost computation as performed in Step 3 included the storage costs just described in the previous section as well as several additional costs such as the .