This preview shows page 1. Sign up to view the full content.
Unformatted text preview: TLB miss NOTE: units are all ns EXCEPT disk access is ms. Problem 2: a. Write through should be used to maintain consistency. If some processors have a value cached, and one processor changes that value, the other processors would not know about that value. With a write through cache (in addition to a broadcast "invalidate" message) the processors would not be using stale values. b. As mentioned in the answer for part a, the problem is stale data. One way to solve this problem is to invalidate stale entries . A way to prevent the situation would be to require processors to specify whether the values were for writing or reading. Shared reading would be ok; however, when a processor wanted to write, it would have to either wait until the readers were done, or send an invalidate message to those readers. Exclusive access by writers would then be allowed....
View Full Document
- Fall '08