1. CS61C Midterm #2 Review Session A little Cache goes a long way

2. The Ideal Memory System Fast Cheap (Large)

3. Actual Memory Systems Fast, Expensive (Small) Slow, Cheap (Large)

4. Idea: Multilevel Memory (cache) + =

5. The Cache CPU • Store recently used data in fast memory • Cache Hit • Address we’re looking for is in cache • Cache Miss • Not found… read memory and insert into cache • This works because… Tag Data Main Memory

6. Locality Just referenced x address Spatial Locality data Reference to data near x likely Temporal Locality stack Likely to reference x again soon code time

7. Computing Average Access Time Q: Suppose we have a cache with a 5ns access time, main memory with a 60ns access time, and a cache hit rate of 95%. What is the average access time?

8. Cache Design Issues • Associativity • Fully associative, direct-mapped, n-way set associative • Block Size • Replacement Strategy • LRU, etc. • Write Strategy • Write-through, write-back

9. An Example

10. Multiple Choice (1) • LRU is an effective cache replacement policy primarily because programs • exhibit locality of reference • usually have small working sets • read data much more frequently than writing data • can generate addresses that collide in the cache

11. Multiple Choice (2) • Increasing the associativity of a cache improves performance primarily because programs • exhibit locality of reference • usually have small working sets • read data much more frequently than writing data • can generate addresses that collide in the cache

12. Multiple Choice (3) • Increasing the block size of a cache improves performance primarily because programs • exhibit locality of reference • usually have small working sets • read data much more frequently than writing data • can generate addresses that collide in the cache