|
|
|
| Table 5.3 System environment effects on miss rate. |
| | Environments | Miss Rate | | User program only | DTMR | | User program plus operating system, I/O | MR adjusted by Figure 5.22 or Appendix A. | | MP:Multiple user programs and system. Control passes round-robin among programs ("warm" cache). | MR adjusted by Q and MP in Figures 5.25-5.24 or Appendix A. | | T: Transactions or multiple short user (Q instructions) programs plus system. Control always passes to a new user program ("cold" cache). | MR adjusted by Q in Figure 5.26 or Appendix A. |
|
| |
|
|
|
|
A cache that contains several processes among which control passes from time to time is said to be a warm cache. For all but very small caches, each process finds some of its working set present when it is restarted. |
|
|
|
| |
|
|
|
|
A user program together with its required system support is also a warm cache. Available data for caches with low degrees of multiprogramming seem consistent with data for user plus system environment. Even a cold cache, where each user task is new to the system, has some system support working set. |
|
|
|
|
|
| |
|
|
|
|
The effects of the operating system and the degree of multiprogramming on cache performance have not been widely studied. The studies available [4, 54] indicate that operating systems behave rather differently from applications. |
|
|
|
| |
|
|
|
|
Operating system policy can have a significant effect on cache performance. The working set of the run-time systems kernel will obviously affect the cache miss rate. The cold and warm effects outlined here are direct functions of the degree of multiprogramming. Environments of many short transactions that run through to completion approximate a cold cache with a high degree of multiprogramming. For such environments, increasing the cache size beyond a relatively modest capacity has little or no effect on performance. Such applications truly tax the memory system designer. |
|
|
|
|
|
|
|