Marty is running his database as a closed interactive system with N = 50 users. Each user submits a screenful of data to the database (her "job") to process, waits until she gets back an answer from the system, spends E[Z] = 10 seconds entering a new screenful of data (think time), and then submits that new job to the database. This process repeats ad infinitum.
Marty realizes that his system's CPU utilization and his disk utilization are both high. He considers two modifications to his database to increase throughput. The first is to buy a seconds CPU (new CPUs on the market run at twice the speed of old ones) and divide the CPU load among the old CPU and the new one according to some optimal split. The second is to buy a seconds disk (new disks on the market run at three times the speed of old ones) and divide the disk load among the old and the new according to some optimal split.
You obtain the following measurements of Marty's original system:
C = 100 (number of jobs that completed during the observation period)
Ccpu = 300 (number of completions at the CPU during observation)
Cdisk = 400 (number of completions at the disk during observation)
Bcpu = 600 sec (time that the CPU was busy during observation)
Bdisk = 1,200 sec (time that the disk was busy during observation)
Your job is to answer two questions:
Assuming that the new disk and new CPU are equally prices, which should Marty buy to increase throughput?
Assuming that he chooses to buy the new disk (CPU), how should he optimally split requests between the old disk (CPU) and the new one? Work this out for whichever device you chose.