Lightweight Determinism in Large-Scale Networks

Citation Author(s):
Andrea
Francini
Raymond
Miller
Bruce
Cilli
Charles
Payette
Catello
Di Martino
Submitted by:
Andrea Francini
Last updated:
Fri, 02/09/2024 - 16:46
DOI:
10.21227/af98-p846
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

The data set was used to produce three figures (Figs. 4, 5, and 6) in the experimental section of a manuscript submitted for review to the IEEE Communications Magazine. The submission, titled Lightweight Determinism in Large-Scale Networks, describes a novel approach to the realization of network determinism in packet-switched networks of generic size and topology. The approach builds on a novel per-flow traffic shaper that controls the ingress links of the deterministic network domain, where transmission opportunities are distributed to packet queues based not only on their guaranteed service rates, but also on awareness of the network paths over which the respective flows are routed.

The first figure (Fig. 4) shows the distribution of maximum-latency samples collected in a proof-of-concept system that reproduces a small data-center fabric with Clos topology, using three different versions of the ingress-link scheduler: a single FIFO queue for all deterministic flows, a conventional per-flow shaper, called FLAT, and the new per-flow shaper of the submission, called the routing-aware shaper (RAS). In five different runs per shaper configuration, the network load is set at 9.6%, 24%, 48%, 72%, and 96%.

The second figure (Fig. 5) shows results from two runs of the experiment with 82% load and FLAT and RAS shapers. The plots compare the distributions of the maximum latency samples measured with the two shapers, and the distributions of service opportunities that each shaper grants to the traffic aggregates that it dispatches to the three interior links of the network. The differences between the distributions of service opportunities explain the large difference between the distributions of latency samples produced by the two shapers.

The third figure (Fig. 6) repeats the same types of plots as Fig. 5 for a traffic scenario with identical load (82%) but uneven distribution of the data rates among the 3840 flows of the scenario (ranging from 5 Mb/s to 29 Mb/s). Code for generation of the periodic sequences of service opportunities can be found here: GitHub - nokia/PSS-generator .

Instructions: 

The three README files refer to the figures of the three data sets. The names are derived from the placement of the three figures within the submission (Figs. 4, 5, and 6).