| With the popularity of digital products, the total amount and type of family picturesgrow explosively which results the storage and indexing needs for the massive images.The design of an efficient storage management and retrieval system for home massivepersonal picture file to become an urgent problem. Therefore, massive picture storagemanagement and retrieval prototype system is designed and implemented.The system based on C/S model, contains data upload, semantic extraction, efficientretrieval mechanism and optimization techniques. Data upload is designed to use FTPtransfer image files in the user’s photographic equipment to the backend server; Semanticextract is implemented by adding extended attributes (such as location, time, etc.) for theuploaded picture from client; In server memory, key-value retrieval mechanism is buildbased on hierarchical index. In inserting, using Bloom Filter build retrieval set, then thekey is hashed to get the address of AVL Tree of the second layer, and put <key, value> intoAVL Tree; In querying, through the Bloom Filter of the first layer filter the query, then thequery is hashed to get the address of AVL Tree of the second layer, and query result fromAVL Tree; Server provide remote calls, include insert, delete, update, query operations.Finally, by combining appending information to log file on disk with snapshot technology,all memory index information is saved to disk log file in order to protect the security ofthe memory index information.Experimental results show that the back-end server memory index structure can write48600key-value pairs, and read out of about377,800a key-value pair per second. For adirectory with140,000files, by using the command ‘find’ of Linux File System, it takes0.5seconds to find a file. Assume that each file has10properties, there will be1400000key-value pairs, it takes30.78seconds to insert them into the memory index structure.And it just takes30microseconds to find a file by the index structure. |
PDF Full Text Request