Improving Cache System Performance Based On Access Patterns

With the rapid development of mobile Internet technology,the number of people and devices connected to the network,the scale of data in the network has exploded.Cache technology is an important technology in computer system.However,in the current cache system,the dependency relation and access patterns between the data items in the access sequence are not used well,especially in the network cache system,which leads to suboptimal cache utilization.First,the caching strategy of the current cache server doesn`t know which data will not be used in the future,and which data will be used frequently later in the service.Multiple access patterns from visitors make the server more vulnerable and reduce the whole system benefit from caching.In this paper,a cache management strategy is proposed to solve the above problems.It extracts the set of frequent items from the access stream,according to the spatial and temporal characteristics of each data in the access stream,the cache management is guided by the evaluation results of the dependencies between the data items.Experiments show that compared with LRU algorithm in traditional system,the hit rate of load generation is improved more than twice under experimental conditions,and the hit rate of real load is also improved about twice.On the other hand,the widely used data placement method in cluster also lacks the ability to be aware of the access service patterns.Firstly,the hash algorithms have some defects in consistency and the data skew of the Zipfian distribution that the workload obeys.In this case,the high-frequency access pattern of two or more related data are mapped to the same physical machine.High-frequency access requests may cause the access hotspot skew in the cluster,affecting the performance of the system.To solve this problem,we introduce the idea of awareness of access patterns into the data migration of Redis cluster.A migration strategy is designed to distribute the data items within the same access pattern in different physical machines,which makes each physical node in the cluster load the user's access request to these high-frequency reused data items.Experimental results show that this method can effectively reduce cluster imbalance and improve cluster throughput in Redis cluster environment.The above research can be applied and not limited to Web static content caching and cache systems such as K-V storage,distributed data store,etc.