Cache coherence hinder the normal flow of work by reducing down the speed. Let’s discuss what is cache coherence problem to overcome it.
Cache coherence happens when two caches or more mirror the same resource. In case every cache agrees on the content of the cache, the caches become coherent. When at least a single cache gets changed, the caches become incoherent unless the changes get propagated to any other cache.
One effective way of dealing with Cache coherence is to work with PAFS. This is one file system which is a cooperative cache (PG-LRU). These cooperative caches do not require any coherence mechanism. If you use this kind of file system, there will not be any loss in the performance level if you compare the same with any other present cooperative cache algorithms.
There have been lots of researches in recent times, which involve distributed/parallel and parallel I/O file systems. There is a common aspect among the maximum number of the file systems is a notion that there should be the cooperation between different nodes for achieving better performance for the system. Every node in cooperative caches works together to make one global cache. The type of cooperation enhances the size of the cache. Thus, it also improves the ratio of hits, which improves the performance of the system eventually.
image-source
An important observation is that traditionally a lot of importance is given on the ratio of local hits. It maintains a high standard for the same. Modern day advancements have shown that the ratio of local hits is not that important as it might have been believed to be. When you implement remote hits on an efficient level, the high ratio of local hits certainly becomes secondary.
One more critical result shows that global hit ratios do not benefit much regarding cooperative caches. There is every possibility that such ratios may even worsen the same. On the other hand, there is a possibility of extending the same to a pre-fetching area. In case a block has to be overwritten, no pre-fetching ought to be done.
There can be conclusive notions that gigantic global caches are offered by cooperative mechanisms. These global caches have to be used through forceful pre-fetching algorithms for improving the overall performance of the file system. Pre-fetching algorithms which fall in such a category must also be presented.
There are also results which are valid with interconnection networks which are fast. However, these may not work at their best with interconnection networks which extremely fast. It should run well on networks which has ten times less bandwidth than its local bandwidth memory.
