UC Santa Cruz

In-network computing is a growing area of research where computations are more tightly integrated with the networking substrate and placed dynamically within the network at appropriate locations. However, this yield new issues in the domain of measurements and networking: it adds a dimension in the measurement domain, namely that of the computation for some tasks. We discuss MECANO, an architecture that integrates application measurement with network measurement to optimize the placement of workloads in the network. We have implemented an initial proof of concept of this architecture and validated its performance in our testbed. In particular, keeping track at the network layer of the location of running instances of a specific function can significantly speed up the completion time for a specific client request.