Integrity verification mechanisms are widely used in non-volatile memory systems to ensure data security,but they will introduce extra metadata.Metadata needs to be stored correctly to avoid that data cannot be used after the system crashes.SIT(SGX Integrity Tree)is efficient because of parallel computing between layers among the above mechanisms.However,to ensure the crash consistency of SIT,the existing works need to update multiple nodes for each write operation and persist the corresponding metadata,causing high write latency and seriously reducing system performance,as well as long crash recovery time.For the systems that require fast recovery after crashes,we propose a scheme named FUAR for fast update and recovery of secure metadata.FUAR accelerates crash recovery by reducing the number of nodes that need to be updated and the metadata that must be persisted for each data write operation.FUAR introduces the Ptable(Persist table)in the non-volatile memory controller,and records the update information of the SIT leaf node in the Ptable during the write operation,without updating the SIT intermediate node.When the Ptable is full,the SIT is updated with the recorded information,and the dirty nodes of the SIT are written to non-volatile memory.After the system crashes,the update information recorded in the Ptable can be used to quickly infer the incorrect nodes in the SIT,and complete the node recovery.For the systems that require higher performance,we propose a secure metadata write traffic optimization scheme WTOP.WTOP decreases write traffic by reducing the persistence of dirty nodes at the cost of a slight increase in recovery time.WTOP allocates a backup area in non-volatile memory.When the Ptable is full,WTOP first updates the SIT and then writes the Ptable to the backup area.When the backup area is full,the dirty nodes corresponding to the replaced Ptable should be persisted.When the dirty node is persisted,the time information is recorded for the node,so as to ensure that the system can correctly select the update information of the backup area to restore the node after the system crashes.Experimental results show that compared with schemes without crash consistency,FUAR and WTOP only cause 5.24% and 2.52% performance loss when crash consistency is guaranteed.Compared with the state-of-the-art work,the performance loss of FUAR and WTOP is reduced by 73.96% and 87.48%,and the write latency is reduced by 30.43% and38.59%,respectively.In addition,WTOP reduces write traffic by up to 43.78% based on FUAR.In terms of recovery time,FUAR is about one percent of the state-of-the-art work,and WTOP’s recovery time is comparable to the state-of-the-art work. |