Skip to main content

Trading Latency for Compute

Introducing compute hierarchy in the network boosts in-network computing

Diagram of two pyramids, with text "CPU Hierarchy vs network hierarchy"

In-network computing is known to be bounded by switch resources. In a new work, published in NAI'20, researchers from the group, USI, Yale and Western Digital propose a new heterogeneous approach to in-network computing that extends the capabilities of programmable switch ASICs with FPGAs. The work further presents a proof-of-concept data deduplication application.
The paper is available here.