Accelerating Applications with In-network Computing

Summary:
In-network computing enables offloading of applications traditionally designed to run on general purpose CPUs to high-performance programmable network devices. However, even cutting-edge
programmable network devices do not yet support running more than one application by sharing
the same hardware with other non-dedicated instances. This research project aims to explore
virtualization and security capabilities required to enable running multiple applications on a
programmable network device and how in-network computing can improve the scalability,
availability, and reliability in the cloud.
Description:
Programmable network devices are the key to realize the next generation of network services, largescale distributed computing systems, software-defined networking, cloud and data centre
applications, and Internet of things (IoT). However, several barriers of programmable network
devices hinder this promising in-network computing technique. Each device is exclusively operated
by a single program once configured. As a result, this prevents an efficient sharing the underlying
hardware resources between different applications in a multi-tenant environment where multiple
users or organizations share the same computing infrastructure. In addition, it is hardly possible to
dynamically reconfigure a programmable network device to run a new application without
interrupting its operation. Virtualization is one solution to these problems. It enables multiple
different in-network applications to co-exist on the same physical device and operate in a manner
transparent to other applications. Recent research, MTPSA1
, proposed a methodology for providing
resource, performance, and security isolation are fundamental to support virtualization in
programmable network devices. There are three primary aims of this study: 1) This project aims to
further explore the MTPSA roles and permissions required to support running applications of
mutually untrusted users on programmable network devices. 2) This research project also aims to
quantify the scalability, availability and reliability improvements provided by offloading distributed
applications to high-performance programmable network devices. 3) This study aims to identify and
propose methodological approach for internet service providers (ISPs) to leverage the
programmability of next-generation network devices to facilitate Network-Platform-as-a-Service
models.
In particular, this project has the following objectives:
-Identify a standard set of roles and permissions on multi-tenant programmable switches
required to enable user programs to run in a safe and secure manner.
-Quantify the scalability, availability and reliability improvements that can be provided to
distributed systems in cloud environments using in-network computing.
-Propose ways for internet service providers (ISPs) to leverage the programmability of nextgeneration network devices to facilitate Network-Platform-as-a-Service models.
The novelty of the research lies in the combination of virtualization techniques provided by modern
operating systems, with programmable network hardware that enables in-network computing. As
cloud computing environment are often shared by mutually untrusted applications that are co1 https://eng.ox.ac.uk/computing/projects/in-network-computing/mtpsa/
located in the same network infrastructure, this research project focuses on the security,
performance and resource isolation of in-network applications.
Potential impact:
In-network computing has known benefits in terms of power efficiency, performance and scalability.
Using it to facilitate Network-Platform-as-a-Service models will be a big step towards Net-Zero digital
infrastructure and increase the resilience and availability of critical digital services.
Alignment to EPSRC's strategies and research areas:
This project falls within the EPSRC Information and Communication Technologies (ICT) research