This preview shows page 1. Sign up to view the full content.
Unformatted text preview: as the disadvantage that every miss results in a block transfer from memory to cache. The alternative, known as no-write-allocate, bypasses the cache and writes the word directly to memory. Write-through caches are typically no-write-allocate. Write-back caches are typically write-allocate. Optimizing caches for writes is a subtle and difﬁcult issue, and we are only touching the surface here. The details vary from system to system and are often proprietary and poorly documented. To the programmer trying to write reasonably cache-friendly programs, we suggest adopting a mental model that assumes writeback write-allocate caches. There are several reasons for this suggestion. As a rule, caches at lower levels of the memory hierarchy are more likely to use write-back instead of write-through because of the larger transfer times. For example, virtual memory systems (which use main memory as a cache for the blocks stored on disk) use write-back exclusively. But as logic densities increase, the increased complexity of write-back is becoming less of an impediment and we are seeing write-back caches at all levels of modern systems. So this assumption matches current trends. Another reason for assuming a write-back w...
View Full Document
- Spring '10
- The American