The software-defined networking is used extensively in data centers that provide centralized control for the widely deployed networking resources. The traffic is shaped by rules created by the controller dynamically without modifying the individual switch. The key component that stores rules which are used to process the flows is the flow table which resides in the ternary content addressable memory. The current commercial OpenFlow appliances accommodate limited entries up to 8000 due to its high cost and high power consumption. There are two issues to be considered, where (1) flow table's inability to provide rules during flow table overflow leads to dropping of incoming packets and (2) the significant amount of rule replacement occurs when the traffic in data centers increases which creates massive route requests to controller creating overhead. The proposed scheme prevents flow table overflow using the robust machine learning algorithm called decision tree (Iterative Dichotomiser 3) that allows the flow table to learn its high prioritized fine-grained entries by means of multiple matching attributes. The entries are classified, and the usual eviction process is replaced by pushing the low important entries into counting bloom filter which acts as a cache to prevent flow entry miss. The simulations were carried out using real-time network traffic datasets, and the comparisons with the various existing schemes prove that the proposed approach reduces 99.99% of the controller's overhead and the entries are minimized to 99% providing extra space for new flows.

中文翻译：

---

基于决策树的条目减少方案使用多匹配属性防止SDN环境中的流表溢出

软件定义网络广泛用于数据中心，为广泛部署的网络资源提供集中控制。流量由控制器动态创建的规则调整，无需修改单个交换机。存储用于处理流的规则的关键组件是驻留在三元内容可寻址存储器中的流表。由于其高成本和高功耗，当前的商用 OpenFlow 设备最多只能容纳 8000 个条目。有两个问题需要考虑，其中（1）流表' 在流表溢出期间无法提供规则会导致传入数据包的丢弃，并且 (2) 当数据中心的流量增加时会发生大量规则替换，这会向控制器创建大量路由请求，从而产生开销。所提出的方案使用称为决策树（迭代二分法 3）的稳健机器学习算法来防止流表溢出，该算法允许流表通过多个匹配属性来学习其高优先级的细粒度条目。条目被分类，通过将低重要条目推送到计数布隆过滤器来代替通常的驱逐过程，该过滤器充当缓存以防止流条目未命中。模拟是使用实时网络流量数据集进行的，