Overview

The execution time of a program is often estimated using the average number of clocks per instruction, also called CPI, where the length of a clock cycle can be estimated as a machine parameter. Typically, the contributions of the CPU and the memory are separated when talking about average CPI. That is, we typically write CPI as the sum of CPI associated with the cpu and the CPI associated with memory effects.

These notes give a limited overview of cache organization and the impact of the memory hierarchy on execution time or CPI calculations.

Don't forget to to check out the worksheet, and the book for more details. Note that this document adopts the verbiage appropriate for logical memory addresses. That is, the notion of cache management presented here is technically accurate; however, it applies to the one-user one-program world of computing, prior to the use of virtual memory to allow each user the illusion of full processor control and resource access. Thus, you will need to know this material for management of the memory hierarchy using the viewpoint of the CPU, which is the logical address space in terms of block address, block offset, index bits, and tag bits. You will also know about the impact of virtual memory where physical addresses are used, in terms of page address, page offset, index bits and tag bits, but the rules and concepts will be similar to those for caches.