Interview
The Network as a Program with Nate Foster
- In-network computing is expected to become inevitable over the next several years despite current controversy.
- Increasing latency and flat support trends may render commodity router multicast capabilities problematic if conditions persist for several years.
- Hybrid networks combining packet switching and scheduled crossbars are anticipated as a potential direction within a few years.
- Formal methods for network verification are predicted to become commonplace to facilitate exploration of larger design steps beyond bug prevention.
- The current system may be extended to support richer policy abstractions in the future.
- Tools designed to minimize disruption when rerouting traffic away from hotspots remain in the research phase and are not yet fully implemented.
- Academic developments in routing algebras could eventually influence industry work to create richer compilers.
- The "network as a program" vision is identified as the general trend, though the implementation path is less smooth than originally predicted.
- The use of centralized programs to specify network behavior is expected to continue gaining traction among hyperscalers and large organizations.
- Multicast may prove more useful and underappreciated in contexts beyond trading, such as Twitter's distribution systems.