面向多媒体SoC的存储体访存负载均衡划分方法

发布时间：2018-06-24 04:33

本文选题：内存分配 + 片上计算机系统　；参考：《计算机辅助设计与图形学学报》2015年03期

【摘要】：随着多媒体So C中具备密集访存能力的设备数量增加,设备之间频繁争抢存储体资源,严重影响访存性能.为此提出一种面向多媒体So C的存储体访存负载均衡划分方法.通过操作系统对物理内存的管理,将设备所访问的数据映射到独立的存储体中,避免争抢频繁的设备共享存储体,减少设备间的访存冲突;划分过程基于数据量、延迟分析设备访存行为与访存冲突之间的关系,并以此来均衡各存储体的访问负载,同时提升多个设备的访存性能.该方法不依赖特殊硬件也无需修改上层应用,提供了一种透明的纯软件优化手段.将文中方法应用于真实的多媒体So C的实验结果表明,与基于带宽优先的划分方法相比,该方法在提高带宽利用率的同时降低访存延迟,将解码帧率提升8.4%~12.3%;并且在保证服务质量的情况下,可以通过进一步降低内存工作频率来减少系统功耗.
[Abstract]:With the increase of the number of devices with intensive memory access capacity in multimedia so C, storage resources are frequently scrambled among devices, which seriously affects memory access performance. In this paper, a partition method of storage access load balance for multimedia so C is proposed. Through the management of the physical memory in the operating system, the data accessed by the device is mapped to the independent storage body, so as to avoid the frequent competition for the shared storage of the device, and reduce the memory access conflict between the devices, and the partition process is based on the amount of data. Delay analysis of the relationship between memory access behavior and memory access conflict is used to balance the access load of each storage system and to improve the access performance of multiple devices at the same time. This method does not depend on the special hardware and does not need to modify the upper layer application, and provides a transparent pure software optimization method. The experimental results show that compared with the bandwidth-first partitioning method, the proposed method not only improves the bandwidth utilization but also reduces the memory access delay. The decoded frame rate is increased by 8.4 / 12.3 and the system power consumption can be reduced by further reducing the working frequency of memory under the condition of guaranteeing the quality of service.
【作者单位】：北京大学微处理器研发中心;首都师范大学信息工程学院;
【基金】：国家自然科学基金(61402302)
【分类号】：TP333

【相似文献】