High Level System Design
The COSMOS platform's technical architecture is designed to meet both quantitative and qualitative requirements associated with a fully programmable city-scale shared multiuser advanced wireless testbed facility. In addition to realizing Gbps+ radio speeds with low latency, the design should support a number of capabilities including remote multiuser access, virtualization of resources, open programmability, flexible topology, node mobility, diversity of radio environments, reproducibility, usability, instrumentation, extensibility and interoperability.
Open APIs and full programmability across all the technology components and protocol layers is the most important aspect of the design. Our approach to realizing full programmability is based on a multi-level SDR and network architecture in which signal and protocol processing functions can be flexibly placed at radio nodes, edge cloud resources or general purpose cloud servers depending on the desired functionality. The testbed's focus on ultra-high bandwidth wireless implies the need for significant SDR computing capability in the radio access network - rather than placing all the compute functionality at the radio node, having a second layer of "cloud RAN" capability makes it possible to offload a significant part of the node's function to an infrastructure-based computing cluster. The same edge cloud can be used for network and application level processing particularly in scenarios requiring low latency end to-end response.

Above figure gives a block diagram of the COSMOS testbed showing the three levels of software processing, namely the SDR radio nodes, the edge cloud servers and the general purpose cloud. The diagram also shows the use of a custom all-optical front-haul network fabric to provide an ultra-fast low latency interconnect between radio nodes and edge cloud resources, a key design detail necessary to support both real-time SDR signal processing and time-critical applications. The use of passive WDM technology with fast laser tuning makes it possible to implement switching with 100G bandwidth and latencies under 100 microsec. The front-haul switching is further connected to a more conventional high-speed SDN-based optical switching backhaul network which links the radio nodes and edge cloud sites to a private wide-area distribution network connected to the COSMOS control center, Internet and Internet2 points-of-presence, and general purpose cloud racks (such as the GENI rack or CloudLab cluster at Rutgers). The COSMOS control center (to be located at Rutgers) includes servers for hosting the user portal, scheduling/virtualization services, experiment management services, measurement services and data repository, etc. An experimenter first enters the system through the COSMOS portal and then uses the provided software tools to set up and run an experiment. The system diagram also shows how the wide-area distribution network connects users to other resources including COSMOS sandboxes for early experimentation and debug, as well as to other testbeds such as GENI, ORBIT and CloudLab.