Providing Monitoring Data to Data Centers
University of Calgary
There are thousands of switches and links connected to cloud providers' networks. In such vast and complex networks, failures in hardware, software, and configuration are inevitable. Network performance anomalies, congestion, and packet drops are vital to data centers. The ability to identify the cause of silent failures is becoming increasingly important. Traditional network monitoring tools, such as TCPDUMP, SNAP, and SNMP, have a high overhead on the network, which prevents them from being applicable in today's networks. A high-performance monitoring system should be able to provide a detailed view of the network. The granularity of such a monitoring system should be at the flow level rather than the traditional interface or device level. We intend to utilize the programmable data-plane capability in order to propose a network monitoring system with minimal overhead. Although traditional network monitoring tools are currently being used in data centers, they have a high overhead on the network, making them unsuitable for today's networks. A high-performance monitoring system should be able to provide a detailed view of the network. It is our intention to propose a network monitoring system that leverages the programmable data-plane capability to deliver minimal overhead and provide granular monitoring at the flow level instead of the traditional interface or device level.