| Graph is a universal data structure. It has powerful capabilities in terms ofdata presentations in many situations. More and more applications poweredby graph are on the rise. With the emerging of social network and semanticweb, we are facing challenging from the storage and querying of massivegraph data. In storage aspect, traditional approaches based on disk ordistributed file systems failed to take into consideration the random accessnature of graph applications, thus leading to a big IO threat to intensivevisiting of graph applications. In the querying aspect, traditional algorithmsonly work in relatively small datasets. When the size of the dataset grows,these algorithms cannot accomplish real time queries. Combining our priorwork, we propose a graph storage and modeling solution based on ourdistributed memory cloud: Trinity. Taking the characters of distributedmemory cloud into consideration, we design and implemented a completequerying scheme, which includes work in query language, query execution,and query optimization. Considering the memory limitation in most realscenarios, we further propose a hybrid storage solution which combines ourmemory cloud with RDBMS, and revised query execution and query optimization accordingly. Our evaluation on our largest real dataset showsthat our solution delivers exceptional performance in all aspects. |
