In-Network Computing

In-Network Computing is the offloading of standard applications to run within network devices. In this emerging research area, user-generated data is terminated before it gets to the host, saving CPU cycles and freeing up resources.

In-Network Computing has been applied to date to a range of applications, including caching, measurements, network services, distributed systems and machine learning.

Our research has demonstrated that in-network computing can provide x10,000 throughput improvement, x100 latency reduction and x1,000 power saving per operation.

Some of the research questions we are exploring include:

• How can we exploit commercially-available network device to run applications in the network?
• What network-hardware and programmable-hardware architectures are most suitable for in-network computing?
• Can we design new in-network computing algorithm?

Research of in-network computing interacts with the group’s other research areas: building upon programmable devices, running within networked systems, and enabling scalable and sustainable computing infrastructure.