On Packet Reordering in Time-Sensitive Networks

Abstract

In time-sensitive networks (as in the context of IEEE TSN and IETF DetNet), some limited amount of packet reordering may be acceptable. Re-sequencing buffers are then used to provide in-order delivery of packets; they depend on critical parameters (timeout value and buffer size) and may affect the worst-case delay and the delay jitter. However, there is no precise understanding of per-flow reordering metrics nor of the dimensioning of re-sequencing buffers in order to provide worst-case guarantees, as required in time-sensitive networks. We close this gap as follows. First, we show that a previously proposed per-flow reordering metric, namely, reordering late time offset (RTO) determines the timeout value. If the network can be assumed to be lossless, another previously defined metric, the reordering byte offset (RBO), determines the required buffer size. However, if packet losses cannot be ignored, the required buffer may be larger than the RBO, and the exact value depends on a combination of the delay jitter between the source and the re-sequencing buffer, an arrival curve of the flow at its source, and the timeout. Then we develop a method to compute the RTO for a flow path; the method uses a novel relation with delay jitter and arrival curve, together with a decomposition of the path into non order-preserving and order-preserving elements. We also analyze the effect of re-sequencing buffers on worst-case delay, delay jitter and propagation of arrival curves. We show in particular that, in a lossless (but non order-preserving) network, re-sequencing is "for free", in the sense that it does not increase worst-case delay nor delay jitter, whereas in a lossy network, re-sequencing increases the worst-case delay and delay jitter. We apply the analysis to evaluate the performance impact of placing re-sequencing buffers at intermediate points and illustrate the results on two industrial test cases.

Worst-case Delay Bounds in Time-Sensitive Networks with Packet Replication and Elimination

“…Additionally, we assume that the sequence of incoming data units cannot be incomplete, i.e., for each data unit of the aggregate m ∈ F , m is not lost for the POF. Without the above assumption, the POF algorithm might enter a recovery situation [29], that has been analyzed in [28] and is out of the scope of this paper. REG: Consider a set of flows F crossing n such that, for each flow f ∈ F , o is a diamond ancestor of n in G(f ).…”

“…As described in Sec. II, it still has an effect on the performances of time-sensitive networks [28].…”

“…Two metrics are of interest when quantifying mis-ordering in time-sensitive networks: the reordering late time offset (RTO) and the reordering byte offset (RBO) [28], [31]. In this paper we focus on the mis-ordering as a consequence of the redundancy.…”

“…For a flow f and two vertices n [resp., o] containing the observation points v [resp., w] such that f is packetized at v, w, n is not an EP-vertex of G(f ) and o is a diamond ancestor of n in G(f ), we denote by λ v (f, w) the RTO of the data units of flow f at the observation point v, with respect to their order at w, as defined in [28], [31]. With the restrictions on v and w, λ v (f, w) is well defined from [28], [31] because each data unit of f is observed at most once at w and v, thus the arrival instant of each data unit at w and v is well defined. Similarly, with v and w meeting the same conditions, we denote by π v (f, w) the RBO, as defined in [28], [31] of the data units of flow f at v with respect to their order at the reference w.…”

“…With the restrictions on v and w, λ v (f, w) is well defined from [28], [31] because each data unit of f is observed at most once at w and v, thus the arrival instant of each data unit at w and v is well defined. Similarly, with v and w meeting the same conditions, we denote by π v (f, w) the RBO, as defined in [28], [31] of the data units of flow f at v with respect to their order at the reference w.…”