Another advantage of non-inclusive caches is that they are able to store larger amounts of data, since the data in L1 are not replicated to the lower levels. For those examples, what are the best replacement policies? However, there is a number of cases on which LRU is not efficient. For example: i. It is clear that every reference results a miss, making this replacement policy not optimal. For example, if in this example we use the Most Recently User MRU replacement policy we will get the following figure 3 , which has significantly less misses.
Figure 3: LRU counter-example 1. Figure 5: LRU counter-example 2. The modification is that it checks if the least recently used reference has been referenced again while it was in the cache indicated by a corresponding reference bit.
If the bit is not set, it is evicted. Otherwise, the referenced bit is cleared, and this reference is inserted at the back of the queue it gets a second chance , as if it were a new reference, and this process is repeated. This access pattern results 6 misses when using LRU. On the other hand, if we use the FIFO replacement algorithm as it is depicted in figure 7, we get 4 misses. Figure 7: LRU counter-example 3. As conclusion we have to say that LRU is a good and very intuitive replacement algorithm.
However, we can almost always invent a counter example an access pattern that may cause problems to it. The main problem of LRU is not it has expensive implementation may require special hardware , and not its bad performance in same rare cases. Answer Although trace-driving simulation is a very convenient simulation technique used for the prediction of the performance of computer systems, it has a number of shortcomings [4, 5].
It is very difficult to create such a trace since that would most probably require modifications in the kernel of the operating system. This has as result most of the trace-driving cache simulations to ignore the effect of the interaction of the operating system. Studies have shown that the operating system can significantly modify the executed instruction mix, and produce different ranges of reference locality.
That is, it is really difficult to simulate the executions of meaningful usually big and time-consuming applications based on its trace, since it is difficult to obtain, store and process their trace.
References: [1] Hennesy, J. Computer Architecture: A Quantitative Approach, third edition. Morgan Kaufman Publishers, Inc.
On the inclusion properties for multi-level cache hierarchies. A5: Fixed number of digits after or before the radix point in knows as fixed-point number representation. It is a real data type for a number in computing which can be compared to the more complicated floating point number representation. Follow us on Facebook and Support us with your Like. Vote count: No votes so far! Be the first to rate this post. Tags ACA Notes aca notes vtu advanced computer architecture advanced computer architecture notes advanced computer architecture notes pdf advanced computer architecture pdf.
Your email address will not be published. How useful was this post? Directory Based Cache Coherence. Memory Consistency Model. Data Level Parallelism. Vector Architecture. Vector Processing. GPU Architecture. Memory Hierarchy Design. Warehouse-scale Computers. View code. Programs project Reports Research papers Articles.
0コメント