基于Kubernetes的多云网络成本优化模型

以Kubernetes为代表的云原生编排系统在多云环境中被云租户广泛使用，随之而来的网络观测性问题愈发突出，跨云跨地区的网络流量成本尤为突出。在Kubernetes中引入扩展的伯克利数据包过滤器（extended Berkeleypacketfilter,eBPF）技术采集操作系统内核态的网络数据特征解决网络观测问题，随后将网络数据特征建模为二次分配问题（quadratic assignment problem,QAP），使用启发式搜索与随机搜索组合的方法在实时计算的场景下求得最佳近优解。此模型在网络资源成本优化中优于Kubernetes原生调度器中仅基于计算资源的调度策略，在可控范围内增加了调度链路的复杂度，有效降低了多云多地区部署环境中的网络资源成本。

Cost optimization model for multi-cloud network based on Kubernetes

The cloud-native scheduling system, represented by Kubernetes, is widely used by cloud tenants in a multi-cloud environment.The problem of network observation becomes more and more serious, especially the cost of network traffic across cloud and region.In Kubernetes, the eBPF technology was introduced to collect the network data features of kernel state of operating system to solve the network observation problem, and then the network data features were modeled as QAP, a combination of heuristic and stochastic optimization was used to obtain the best near optimal solution in a real-time computing scenario.This model is superior to the Kubernetes native scheduler in the cost optimization of network resources, which is based on the scheduling strategy of computing resources only, and increases the complexity of scheduling links in a controllable range, effectively reduces the cost of network resources in a multi-cloud area deployment environment.