doi:10.1016/j.peva.2005.07.004
Copyright © 2005 Elsevier B.V. All rights reserved.
“Pay bursts only once” does not hold for non-FIFO guaranteed rate nodes
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Gianluca Rizzo
,
and Jean-Yves Le Boudec
EPFL, CH-1015 Lausanne, Switzerland
Available online 10 August 2005.
Abstract
We consider end-to-end delay bounds in a network of guaranteed rate (GR) nodes. We demonstrate that, contrary to what is generally believed, the existing end-to-end delay bounds apply only to GR nodes that are FIFO per flow. We show this by exhibiting a counter example. Then, we show that the proof of the existing bounds has a subtle, but important, dependency on the FIFO assumption, which was never noticed before. Finally, we give a tight delay bound that is valid in the non-FIFO case; it is noticeably higher that the existing one. In particular, the phenomenon known as “pay bursts only once” does not apply to non-FIFO nodes. These findings are important in the context of differentiated services. Indeed the existing bounds have been applied to cases where a flow (in the sense of the GR definition) is an aggregate of end-user microflows, and it is not generally true that a router is FIFO per aggregate; thus, the GR node model of a differentiated services router cannot always be assumed to be FIFO per flow.
Keywords: Differentiated services; Quality-of-Service; Scheduling algorithm; Network calculus; Guaranteed rate; Aggregate multiplexing
Fig. 1. We consider a (σ,ρ) constrained flow which traverses a succession of M nodes in a network of GR servers. The flow can be interpreted as a differentiated services aggregate flow, in which case (σ,ρ) is the sum of the parameters of the constituent microflows.
Fig. 2. Example of non-FIFO behavior of two GR nodes, traversed by a (σ,ρ) constrained flow, with σ=l, ρ=r=1, and where the propagation delay at all links is zero. ak and bk are, respectively, the arrival times of packets at nodes 1 and 2, and ck are the departure times of packets from node 2. At all nodes only packet 1 takes its maximum possible delay at the node (equal to its GR clock value at the node), whereas all other packets get no delay from the nodes. The end-to-end delay of packet 1 is of 3l−2ε time units: if
, packet 1 gets a larger delay than the delay bound in [11], which is of 2ltime units.
Fig. 3. Evolution of a sequence of packets, at the input to each of three GR nodes on its path, and at the output of node 3. At the input to node 1, there is a sequence with the characteristics described in the proof of Theorem 4.4, and which is (σ,ρ) constrained, with σ=4l,
time units,
time unit. As the propagation delay at all links is of 1 time unit, the delay experimented by the packet marked in black at nodes 1, 2 and 3 (taking into account propagation delay of the link at the output of each node) is, respectively, of 7, 10 and 13 time units (for an end-to-end delay of 30 time units), and the burstiness of the output flow at each node is, respectively, of 7l, 10l and 13l, as predicted by Theorem 4.4.
Table 1.
Symbols used in formulas

Table 2.
Comparison between the values assumed by the FIFO delay bound in Eq. (2) from [11], and by the non-FIFO delay bound in Eq. (9)a
a We made the following assumptions:
M=7nodes; all nodes have the same delay
ea+eb (and the same fixed part of delay
ea); all links introduce the same delay

s; all nodes guarantee a service rate
r=1 Mbit/s to the flow;
ρ=r,
ea=100 ns,
eb=10 ns; all packets have the same length
l=512 bytes.

Corresponding author. Tel.: +41 21 693 5259.