Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MM-INT: Telemetry in Programmable Switches with Multiple Queues using Source-based Multipath Routing

Mateus N. Bragatto, João Paulo M. Clevelares, Cristina K. Dominicini, Rodolfo S. Villaça, Fábio L. Verdi

TL;DR

This work tackles the overhead of telemetry for multi-queue ports in programmable switches by introducing MM-INT, which stores per-queue metadata in device registers and enables a single INT probe to read data from all queues and ports. It combines source-based multicast routing (MPolKA-INT) with a register-based aggregation to eliminate probe duplication and reverse-path overhead, achieving substantial reductions in both probe count and total data moved. Implemented in P4 and evaluated in a Mininet/BMv2 setting, MM-INT demonstrates about a fourfold decrease in probes and roughly a 2.8× reduction in telemetry bytes compared with traditional INT approaches, at the cost of increased per-probe payload and modest switch memory usage. The work identifies practical limitations (MTU fragmentation, uniform queue counts, and hardware adaptation) and outlines future work toward real hardware deployment (e.g., Tofino 2) and scalability enhancements.

Abstract

This article emphasizes the importance of queues associated with the ports of switches in network monitoring. Traditionally, data collection about these queues is done using programmable data planes and telemetry based on INT (In-band Network Telemetry) probes, assuming there is only a single queue per output port. The MM-INT (Multiqueue Multicast - INT) is a solution that utilizes registers to store data from all queues and ports, enabling the efficient collection of monitoring information. The MM-INT avoids probe overload and employs the origin-based routing mechanism and multicast trees for the probes. The results demonstrate significant reductions in the number of probes sent compared to other traditional solutions found in the literature.

MM-INT: Telemetry in Programmable Switches with Multiple Queues using Source-based Multipath Routing

TL;DR

This work tackles the overhead of telemetry for multi-queue ports in programmable switches by introducing MM-INT, which stores per-queue metadata in device registers and enables a single INT probe to read data from all queues and ports. It combines source-based multicast routing (MPolKA-INT) with a register-based aggregation to eliminate probe duplication and reverse-path overhead, achieving substantial reductions in both probe count and total data moved. Implemented in P4 and evaluated in a Mininet/BMv2 setting, MM-INT demonstrates about a fourfold decrease in probes and roughly a 2.8× reduction in telemetry bytes compared with traditional INT approaches, at the cost of increased per-probe payload and modest switch memory usage. The work identifies practical limitations (MTU fragmentation, uniform queue counts, and hardware adaptation) and outlines future work toward real hardware deployment (e.g., Tofino 2) and scalability enhancements.

Abstract

This article emphasizes the importance of queues associated with the ports of switches in network monitoring. Traditionally, data collection about these queues is done using programmable data planes and telemetry based on INT (In-band Network Telemetry) probes, assuming there is only a single queue per output port. The MM-INT (Multiqueue Multicast - INT) is a solution that utilizes registers to store data from all queues and ports, enabling the efficient collection of monitoring information. The MM-INT avoids probe overload and employs the origin-based routing mechanism and multicast trees for the probes. The results demonstrate significant reductions in the number of probes sent compared to other traditional solutions found in the literature.
Paper Structure (13 sections, 7 figures, 2 tables)

This paper contains 13 sections, 7 figures, 2 tables.

Table of Contents

  1. Introdução
  2. Theoretical Background and Research Problem Positioning
  3. Telemetry using INT
  4. MPolKA-INT
  5. Related Work
  6. INT Metadata Collection in Multiple Queues
  7. Detailed Analysis and Discussion of the Solutions
  8. MM-INT Implementation
  9. Experimentation and Evaluation of MM-INT
  10. Functional Evaluation of the MM-INT Implementation
  11. Performance Evaluation of MM-INT
  12. Prototype Limitations
  13. Conclusions and Future Work

Figures (7)

  • Figure 1: Example of a network with In-band Network Telemetry.
  • Figure 2: Multiple queues on an output port of a network device. Adapted from QueueDisciplines.
  • Figure 3: Example of M-PolKA routing mpolka
  • Figure 4: Internal structure of each switch for collecting telemetry metadata from multiple queues.
  • Figure 5: Comparison with Solutions $S1$ and $S2$. Example with 2 queues per port.
  • ...and 2 more figures