摘要
现期等离子体控制系统的控制周期是100微秒,由于数据传输、多CPU间算法匹配失当等原因造成的延迟可以达到80微秒,但理论上的延迟应当不超过15微秒,因此需要平衡多CPU间的负载来提高CPU间的配合度,从而在降低延迟的同时也可以提高控制系统的性能.同时,通过多线程并行技术及共享内存技术将系统中较耗时的读取反射内存卡(Reflective Memory,RFM)过程与进程并行进行,可以将系统的控制周期缩短到50微妙,而延迟也可以降低到20微秒以下,从而极大的提高PCS的工作效率,降低不必要的延迟,避免成为多系统集群中的时间瓶颈.
The control cycle of the Plasma Control System(PCS)currently is 100 μs and the delay could reach to 80 μs due to the data transfer and the inadequate cooperation among different CPUs. However, the delay in theory is within 15μs, so the loads of all the CPUs need to be balanced in order to reduce the delay and improve the performance of this control system. In addition, with the multi-thread parallel technique and shared memory, the time-consuming process of reading data from RFM can run in parallel with the main process, which could shorten the control cycle to 50 μs and the delay to 30 μs or less. Therefore, the efficiency of PCS could be highly promoted and the reduction of unnecessary delay could prevent PCS from becoming the bottleneck of the whole TOKAMAK control system.
出处
《计算机系统应用》
2016年第3期131-135,共5页
Computer Systems & Applications
基金
国家磁约束核聚变能发展专项(2012GB105000
2011GB101000)
国家973重点基础研究发展计划基金(国内配套研究)(2014GB103000)
国家自然科学基金青年科学基金(11205200)
关键词
控制周期
延迟
负载均衡
并行
线程
共享内存
control cycle
delay
load balancing
parallel
thread
shared memory