Infinity: A Scalable Infrastructure for In-Network Applications

TL;DR

Infinity introduces a resource disaggregation approach that moves processing to capable devices within the network, reducing latency and overhead while enhancing scalability and in-network processing capabilities.

Contribution

The paper presents Infinity, a novel infrastructure that enables scalable in-network processing by disaggregating resources and forwarding processing needs with minimal overhead.

Findings

Significantly reduces processing latency in network applications.

Improves scalability of in-network processing systems.

Maintains original data forwarding while enabling processing on capable devices.

Abstract

Network programmability is an area of research both defined by its potential and its current limitations. While programmable hardware enables customization of device operation, tailoring processing to finely tuned objectives, limited resources stifle much of the capability and scalability desired for future technologies. Current solutions to overcome these limitations simply shift the problem, temporarily offloading memory needs or processing to other systems while incurring both round-trip time and complexity costs. To overcome these unnecessary costs, we introduce Infinity, a resource disaggregation method to move processing to capable devices while continuing to forward as the original owner, limiting unnecessary buffering and round-trip processing. By forwarding both the processing need and associated data simultaneously we are able to scale operation with minimal overhead and delay, improving both capability and performance objectives for in-network processing.