InREC: In-network REal Number Computation - Archive ouverte HAL Access content directly
Conference Papers Year :

InREC: In-network REal Number Computation

(1, 2) , (1) , (2) , (3, 2)
1
2
3

Abstract

Current generation of Reconfigurable Match-Action Tables switches are highly programmable, able to support stateful operations and pipeline specifications using languages like P4. Nevertheless, these switches do not offer primitives to support real-valued operations on the data plane, thus requiring support from external servers or middleboxes to perform advanced operations. We introduce InREC, a system that extends the capabilities of programmable switches to support in-network real-valued operations using the IEEE half-precision floating point representation. It relies on decomposing real-valued functions into lookup tables taking into account the RMT model constraints to reach the right trade-off between accuracy and resource usage. InREC prototype on Barefoot Tofino switches demonstrates the efficiency of InREC for in-network computation of different types of operations and its application for in-network logistic regression models used for classification problems. Our evaluation of InREC shows that it is possible to implement complex in-network applications with high accuracy and low latency.
Fichier principal
Vignette du fichier
InREC__In_network_REal_Number_Computation.pdf (1.29 Mo) Télécharger le fichier
Origin : Files produced by the author(s)

Dates and versions

hal-03525052 , version 1 (13-01-2022)

Identifiers

  • HAL Id : hal-03525052 , version 1

Cite

Matthews Jose, Kahina Lazri, Jérôme François, Olivier Festor. InREC: In-network REal Number Computation. IM 2021 - 17th IFIP/IEEE International Symposium on Integrated Network Management, May 2021, Bordeaux / Virtual, France. ⟨hal-03525052⟩
29 View
65 Download

Share

Gmail Facebook Twitter LinkedIn More