Hall of Mirrors
Network Inference of Data Center Architectures (304938)
*Anna Lake, Schneider ElectricKeywords: IoT, Data Center, Network Inference
Data center device architectures (detailing which devices are connected to which) are typically described manually, requiring lengthy on-site assessments. With the implementation of IoT, we now have access to high-volume, live data streams on device activity via sensors and meters, which provide the opportunity for a solution. With the application of network inference, we can summarize generally disorganized data streams in a meaningful way, providing useful feedback to our customers, while minimizing human error, time spent, and cost associated with on-site mappings. We propose a method to map data center architecture. The architecture inferred by this method can be used to identify likely device connections given current conditions (e.g. power and current sensor measurements) all while considering relatively large, staggered polling intervals of device sensors. The architectures can be update easily with this method as new data is made available, improving estimates as the day progresses (micro) while also detecting macro changes to the system over a longer period of time.