Improving Performance through Object Fusion

Technological advancements in semiconductor fabrication, coupled with architectural innovation, have resulted in lightningspeed improvements in processor performance, which doubles every eighteen months. While memory capacity has increased commensurately with processor performance, memory access speeds improve at the glacial rate of almost 10% annually [9]. The memory wall is rising fast, and promotes memory reference behavior to the dominant factor determining overall performance of many important applications (e.g., commercial database systems spend 25% to 50% of their execution time waiting at the memory system [10]). To alleviate this problem we propose object fusion, a compile time technique that co-locates elements of a pointer-based data structure which are likely to be accessed contemporaneously, and re-orders their fields to improve cache line utilization during traversal.