W-Order scan: minimizing cache pollution by application software level cache management for MMDB

摘要: The utilization of shared LLC(Last Level Cache) is important for efficiency multi-core processor. Uncontrolled sharing leads to cache pollution i.e. the weak locality data(single-usage data without re-using) continuously evict strong (frequently re-used data) from LLC in both inner query processing and co-running programs. For analytical MMDB (Main-Memory Database) applications, with skewed star schema DW, more than 95% memory capacity occupied by memory-resident fact table there are only small size dimension tables locality. Cache partitioning must manage different localities inside avoid table. static OS-based suffers insufficient address due large dynamic also movement overhead during re-allocation. In order employ a practical effective policy, we propose an application softwarebased W-order scan policy real application. consecutive physical based proposed reduce misses high controlling page accessing within pages. Another approach page-color index extract bits pages sort bits, when perform scan, can control too supporting OS allocating. We measure L2 miss rate simulating typical hash join operation. experimental results show that DBMSs improve performance through pattern themselves oppose depending on supports hardware or OS.