Research On Load-aware Memory Management Of Stackable Cryptographic File System

With the rapid development of big data technology,the demand for large-scale data storage is also growing.The data store in the device faces potential security threats by unauthorized access during transmission,storage and archives,such as disclosure,theft,and tampering.To avoid unauthorized access,the usage of the stackable cryptographic file system has become one of the solutions to data storage security.The stackable cryptographic file system will maintain plaintext cache and ciphertext cache for crypto level and file system level,resulting in differences exist in the cache architecture of the stackable cryptographic file system and non-stackable file system.However,the stackable cryptographic file system uses the memory management scheme which design and optimize for the non-stackable file system,and cannot efficiently manage the stacked cache in the stackable cryptographic file system.Although the stackable cryptographic file system uses two layers of page cache to retain different forms of cache pages,the same file data is cached multiple times,which causes a serious waste of memory resources.To explores the performance optimization of the stackable cryptographic file systems,this thesis research the memory management mechanism of stackable cryptographic file systems in depth.By comparing the performance indicators of the non-stackable file system and the stackable cryptographic file system,and breakdown the latency of the requests,it is found that the redundant cache caused by the introduction of the crypto level will increase the system overhead and drop the performance,which also shows that the existing memory management mechanism needs a further discussion and research on how to manage the redundant cache.To solve the performance bottleneck caused by the existing memory management scheme,we present Load-aware Memory Management(LMM)scheme for the enterprise-level stackable cryptographic file system.LMM design a load-aware scheduling strategy for the stackable cryptographic file system,which utilize the request type as the basis for the decision of the best page cache,dynamically cache the required pages to the best page cache and eliminate the page in another page cache,achieving the purpose of eliminating redundant cache in the system.Second,LMM design a capable mechanism for non-redundant page conversion to manage the conversion between plaintext pages and ciphertext pages,which is adaptively converted according to workload and the status of the page,and ultimately achieves the purpose of performance improvement.Our evaluation shows that the proposed scheme reduces page faults caused by cache miss up to 54.4% compared with the original memory management scheme in the stacked cryptographic file system,and also get 34.39% performance improvement for system latency.Even for real application-level workload,LMM can also bring significant performance improvements to stackable cryptographic file systems.