PCR*-Tree: PCM-Aware R*-Tree

Phase change memory (PCM) is a byte-addressable type of non-volatile memory. Compared to other volatile and non-volatile memories, PCM is two to four times denser than dynamic random access memory (DRAM). It has better read latency than NAND flash memory. Even though the write endurance of PCM is 10 times better than NAND flash memory, it is still limited to times per PCM cell. When researchers solve the write endurance problem of PCM, it could be used as an alternative to DRAM for in-memory databases and as an efficient main memory or storage system. Decreasing and balancing the number of writes among PCM cells can solve the endurance problem and, where possible, keep PCM cells usable. Many current applications use spatial data such as location information; for this reason, storing spatial data in memory is very important. R-tree is a well-known data structure that can handle spatial data; we propose using its variant, R*-tree, over PCM because it has more compact structure with less overlaps. Our objective is to design a PCR*-tree – a novel PCM-aware R*-tree that can store spatial data. Initially, we examine how R*-tree causes endurance problems in PCM, and then, we optimize it for PCM. Furthermore, the performance of R*-tree is very poor, especially for insertion, which needs to be solved since it will be used for in-memory databases. According to our experimental results, when the benchmark dataset is used, PCR*-tree dramatically reduced the number of write operations to PCM in average 30 times. Moreover, in average the insert performance is improved by more 70%. These results suggest our new method outperforms existing ones for the PCM endurance problem, as well as in its performance.