A study of memory management for web-based applications on multicore processors

• Authors: Hiroshi Inoue;Hideaki Komatsu;Toshio Nakatani
• Venue: Proceedings of the 2009 ACM SIGPLAN conference on Programming language design and implementation
• Year: 2009

More and more server workloads are becoming Web-based. In these Web-based workloads, most of the memory objects are used only during one transaction. We study the effect of the memory management approaches on the performance of such Web-based applications on two modern multicore processors. In particular, using six PHP applications, we compare a general-purpose allocator (the default allocator of the PHP runtime) and a region-based allocator, which can reduce the cost of memory management by not supporting per-object free. The region-based allocator achieves better performance for all workloads on one processor core due to its smaller memory management cost. However, when using eight cores, the region-based allocator suffers from hidden costs of increased bus traffics and the performance is reduced for many workloads by as much as 27.2% compared to the default allocator. This is because the memory bandwidth tends to become a bottleneck in systems with multicore processors. We propose a new memory management approach, defrag-dodging, to maximize the performance of the Web-based workloads on multicore processors. In our approach, we reduce the memory management cost by avoiding defragmentation overhead in the malloc and free functions during a transaction. We found that the transactions in Web-based applications are short enough to ignore heap fragmentation, and hence the costs of the defrag-mentation activities in existing general-purpose allocators outweigh their benefits. By comparing our approach against the region-based approach, we show that a per-object free capability can reduce bus traffic and achieve higher performance on multicore processors. We demonstrate that our defrag-dodging approach improves the performance of all the evaluated applications on both processors by up to 11.4% and 51.5% over the default allocator and the region-based allocator, respectively.