Based on introducing equipments and technology conditions of continuous casting plant of the 4th sub-factory of general steelmaking factory of WISCO, the operation of tundish casting auto-start, control of withdrawal ...Based on introducing equipments and technology conditions of continuous casting plant of the 4th sub-factory of general steelmaking factory of WISCO, the operation of tundish casting auto-start, control of withdrawal straightening device drive start and control of mould level are analyzed, the critical effect of liquid level control is emphasized in this part. Through comprehensive industrial scale experiments, several relevant problems were solved in application process, which promoted the rate of auto-casting of the factory to beyond 98 percent stably from 39 at the beginning.展开更多
在容器技术和微服务框架的普及背景下,无服务器计算为开发者提供了一种无需关注服务器操作以及硬件资源管理的云计算范式.与此同时,无服务器计算通过弹性扩缩容实时地适应动态负载变化,能够有效降低请求响应延时并且减少服务成本,满足...在容器技术和微服务框架的普及背景下,无服务器计算为开发者提供了一种无需关注服务器操作以及硬件资源管理的云计算范式.与此同时,无服务器计算通过弹性扩缩容实时地适应动态负载变化,能够有效降低请求响应延时并且减少服务成本,满足了客户对于云服务成本按需付费的需求.然而,无服务器计算中面临着弹性扩缩容需求导致的冷启动延迟问题.提前预热函数实例能够有效地降低冷启动发生频率和延时.然而,在云环境中流量突发问题极大地增加了预测预热函数实例数的难度.针对上述挑战,提出了一种基于概率分布的弹性伸缩算法(probability distribution based auto-scaling algorithm,PDBAA),利用监控指标历史数据预测未来请求的概率分布,以最小化请求响应延时为目的计算预热函数实例的最佳数量,并且PDBAA能够有效地结合深度学习技术的强大预测功能进一步提升性能.在Knative框架中,通过NASA和WSAL数据集对算法进行了验证,仿真实验表明,相比于Knative弹性伸缩算法以及其他预测算法,所提出的算法弹性性能提升了31%以上,平均响应时间降低了16%以上,能够更好地解决流量突发问题,有效地降低了无服务器计算请求的响应延时.展开更多
文摘Based on introducing equipments and technology conditions of continuous casting plant of the 4th sub-factory of general steelmaking factory of WISCO, the operation of tundish casting auto-start, control of withdrawal straightening device drive start and control of mould level are analyzed, the critical effect of liquid level control is emphasized in this part. Through comprehensive industrial scale experiments, several relevant problems were solved in application process, which promoted the rate of auto-casting of the factory to beyond 98 percent stably from 39 at the beginning.
文摘在容器技术和微服务框架的普及背景下,无服务器计算为开发者提供了一种无需关注服务器操作以及硬件资源管理的云计算范式.与此同时,无服务器计算通过弹性扩缩容实时地适应动态负载变化,能够有效降低请求响应延时并且减少服务成本,满足了客户对于云服务成本按需付费的需求.然而,无服务器计算中面临着弹性扩缩容需求导致的冷启动延迟问题.提前预热函数实例能够有效地降低冷启动发生频率和延时.然而,在云环境中流量突发问题极大地增加了预测预热函数实例数的难度.针对上述挑战,提出了一种基于概率分布的弹性伸缩算法(probability distribution based auto-scaling algorithm,PDBAA),利用监控指标历史数据预测未来请求的概率分布,以最小化请求响应延时为目的计算预热函数实例的最佳数量,并且PDBAA能够有效地结合深度学习技术的强大预测功能进一步提升性能.在Knative框架中,通过NASA和WSAL数据集对算法进行了验证,仿真实验表明,相比于Knative弹性伸缩算法以及其他预测算法,所提出的算法弹性性能提升了31%以上,平均响应时间降低了16%以上,能够更好地解决流量突发问题,有效地降低了无服务器计算请求的响应延时.
