|
|
|
|
|
|
|
Figure 5.14
Relative miss rate with respect to a fully associative cache. We
assume that 8w and higher degrees of associativity behave the same as a
fully associative cache. |
|
|
|
|
| Table 5.2 Associativity adjustment for DTMR. |
| | Unadjusted MR | | | | | | | Direct | | | | | | | 2-way | | | | | | | 4-way | | | | | |
|
|
|
|
|
|
Table 5.2 is an alternative form of the adjustment. Here we find the DTMR rate and use it directly to find the adjustment factor. Table 5.2 basically provides the same overall adjustment as Figure 5.14, but with slightly less resolution and slightly more flexibility. We will use Figure 5.14 when adjusting direct DTMR data, and Table 5.2 when we adjust derived cache datasplit caches, caches with systems effects, etc. |
|
|
|
|
|
|
|
|
For architectures other than the R/M architecture, adjustments can be made for code density of the referenced architecture. We treat these issues in more detail later in this chapter. |
|
|
|
|
|
|
|
|
A Note on Anomalous Results |
|
|
|
| |
|
|
|
|
The adjusted DTMR may give obviously anomalous or impossible results, such as miss rates over 1.0. These situations may develop for small direct mapped caches. Clearly, the miss rate cannot exceed 1.0. Actually, such caches may still be of practical value, but our data indicate that for some applications, the miss rate will be very poor. The designer cannot prudently rely on such caches unless users and applications developers are made aware of the design limitations. |
|
|
|
|
|
|
