| Network Working Group | Z. Sarker |
| Internet-Draft | Ericsson AB |
| Intended status: Informational | X. Zhu |
| Expires: September 14, 2020 | J. Fu |
| Cisco Systems | |
| March 13, 2020 |
Evaluation Test Cases for Interactive Real-Time Media over Wireless Networks
draft-ietf-rmcat-wireless-tests-11
The Real-time Transport Protocol (RTP) is a common transport choice for interactive multimedia communication applications. The performance of these applications typically depends on a well-functioning congestion control algorithm. To ensure a seamless and robust user experience, a well-designed RTP-based congestion control algorithm should work well across all access network types. This document describes test cases for evaluating performances of candidate congestion control algorithms over cellular and Wi-Fi networks.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on September 14, 2020.
Copyright (c) 2020 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
Wireless networks (both cellular and Wi-Fi [IEEE802.11]) are an integral and increasingly more significant part of the Internet. Typical application scenarios for interactive multimedia communication over wireless include from video conferencing calls in a bus or train as well as live media streaming at home. It is well known that the characteristics and technical challenges for supporting multimedia services over wireless are very different from those of providing the same service over a wired network. Although the basic test cases as defined in [I-D.ietf-rmcat-eval-test] have covered many common effects of network impairments for evaluating RTP-based congestion control schemes, they remain to be tested over characteristics and dynamics unique to a given wireless environment. For example, in cellular networks, the base station maintains individual queues per radio bearer per user hence it leads to a different nature of interactions between traffic flows of different users. This contrasts with a typical wired network setting where traffic flows from all users share the same queue at the bottleneck. Furthermore, user mobility patterns in a cellular network differ from those in a Wi-Fi network. Therefore, it is important to evaluate the performance of proposed candidate RTP-based congestion control solutions over cellular mobile networks and over Wi-Fi networks respectively.
The draft [I-D.ietf-rmcat-eval-criteria] provides the guideline for evaluating candidate algorithms and recognizes the importance of testing over wireless access networks. However, it does not describe any specific test cases for performance evaluation of candidate algorithms. This document describes test cases specifically targeting cellular and Wi-Fi networks.
A cellular environment is more complicated than its wireline counterpart since it seeks to provide services in the context of variable available bandwidth, location dependencies and user mobilities at different speeds. In a cellular network, the user may reach the cell edge which may lead to a significant amount of retransmissions to deliver the data from the base station to the destination and vice versa. These radio links will often act as a bottleneck for the rest of the network and will eventually lead to excessive delays or packet drops. An efficient retransmission or link adaptation mechanism can reduce the packet loss probability but there will remain some packet losses and delay variations. Moreover, with increased cell load or handover to a congested cell, congestion in the transport network will become even worse. Besides, there exist certain characteristics that distinguish the cellular network from other wireless access networks such as Wi-Fi. In a cellular network --
Hence, a real-time communication application operating over a cellular network needs to cope with a shared bottleneck link and variable link capacity, events like handover, non-congestion related loss, abrupt changes in bandwidth (both short term and long term) due to handover, network load and bad radio coverage. Even though 3GPP has defined QoS bearers [QoS-3GPP] to ensure high-quality user experience, it is still preferable for real-time applications to behave in an adaptive manner.
Different mobile operators deploy their own cellular networks with their own set of network functionalities and policies. Usually, a mobile operator network includes a range of radio access technologies such as 3G and 4G/LTE. Looking at the specifications of such radio technologies it is evident that only the more recent radio technologies can support the high bandwidth requirements from real-time interactive video applications. The future real-time interactive application will impose even greater demand on cellular network performance which makes 4G (and beyond) radio technologies more suitable for such genre of application.
The key factors in defining test cases for cellular networks are:
However, these factors are typically highly correlated in a cellular network. Therefore, instead of devising separate test cases for individual important events, we have divided the test case into two categories. It should be noted that the goal of the following test cases is to evaluate the performance of candidate algorithms over the radio interface of the cellular network. Hence it is assumed that the radio interface is the bottleneck link between the communicating peers and that the core network does not introduce any extra congestion along the path. Consequently, this draft has kept as out of scope the combination of multiple access technologies involving both cellular and Wi-Fi users. In this latter case the shared bottleneck is likely at the wired backhaul link. These test cases further assume a typical real-time telephony scenario where one real-time session consists of one voice stream and one video stream.
Even though it is possible to carry out tests over operational cellular networks (e.g., LTE/5G), and actually such tests are already available today, these tests cannot in general be carried out in a deterministic fashion to ensure repeatability. The main reason is that these networks are controlled by cellular operators and there exist various amounts of competing traffic in the same cell(s). In practice, it is only in underground mines that one can carry out near deterministic testing. Even there, it is not guaranteed either as workers in the mines may carry with them their personal mobile phones. Furthermore, the underground mining setting may not reflect typical usage patterns in an urban setting. We, therefore, recommend that a cellular network simulator is used for the test cases defined in this document, for example -- the LTE simulator in [NS-3].
The goal of this test is to evaluate the performance of the candidate congestion control algorithm under varying network load. The network load variation is created by adding and removing network users a.k.a. User Equipments (UEs) during the simulation. In this test case, each user/UE in the media session is an endpoint following RTP-based congestion control. User arrivals follow a Poisson distribution proportional to the length of the call, to keep the number of users per cell fairly constant during the evaluation period. At the beginning of the simulation, there should be enough time to warm-up the network. This is to avoid running the evaluation in an empty network where network nodes are having empty buffers, low interference at the beginning of the simulation. This network initialization period should be excluded from the evaluation period. Typically, the evaluation period starts 30 seconds after test initialization.
This test case also includes user mobility and some competing traffic. The latter includes both the same types of flows (with same adaptation algorithms) and different types of flows (with different services and congestion control schemes).
Each mobile user is connected to a fixed user. The connection between the mobile user and fixed user consists of a cellular radio access, an Evolved Packet Core (EPC) and an Internet connection. The mobile user is connected to the EPC using cellular radio access technology which is further connected to the Internet. At the other end, the fixed user is connected to the Internet via wired connection with sufficiently high bandwidth, for instance, 10 Gbps, so that the system bottleneck is on the cellular radio access interface. The wired connection to in this setup does not introduce any network impairments to the test; it only adds 10 ms of one-way propagation delay.
The path from the fixed user to the mobile users is defined as "Downlink" and the path from the mobile users to the fixed user is defined as "Uplink". We assume that only uplink or downlink is congested for mobile users. Hence, we recommend that the uplink and downlink simulations are run separately.
uplink
++))) +-------------------------->
++-+ ((o))
| | / \ +-------+ +------+ +---+
+--+ / \----+ +-----+ +----+ |
/ \ +-------+ +------+ +---+
UE BS EPC Internet fixed
<--------------------------+
downlink
Figure 1: Simulation Topology
The values enclosed within "[ ]" for the following simulation attributes follow the same notion as in [I-D.ietf-rmcat-eval-test]. The desired simulation setup is as follows --
The investigated congestion control algorithms should result in maximum possible network utilization and stability in terms of rate variations, lowest possible end to end frame latency, network latency and Packet Loss Rate (PLR) at different cell load levels.
The goal of this test is to evaluate the performance of candidate congestion control algorithm when users visit part of the network with bad radio coverage. The scenario is created by using a larger cell radius than that in the previous test case. In this test case, each user/UE in the media session is an endpoint following RTP-based congestion control. User arrivals follow a Poisson distribution proportional to the length of the call, to keep the number of users per cell fairly constant during the evaluation period. At the beginning of the simulation, there should be enough amount of time to warm-up the network. This is to avoid running the evaluation in an empty network where network nodes are having empty buffers, low interference at the beginning of the simulation. This network initialization period should be excluded from the evaluation period. Typically, the evaluation period starts 30 seconds after test initialization.
This test case also includes user mobility and some competing traffic. The latter includes the same kind of flows (with same adaptation algorithms).
Same as defined in Section 2.1.1
The desired simulation setup is the same as the Varying Network Load test case defined in Section 2.1 except the following changes:
The investigated congestion control algorithms should result in maximum possible network utilization and stability in terms of rate variations, lowest possible end to end frame latency, network latency and Packet Loss Rate (PLR) at different cell load levels.
The evaluation criteria document [I-D.ietf-rmcat-eval-criteria] defines the metrics to be used to evaluate candidate algorithms. Considering the nature and distinction of cellular networks we recommend that at least the following metrics be used to evaluate the performance of the candidate algorithms:
Given the prevalence of Internet access links over Wi-Fi, it is important to evaluate candidate RTP-based congestion control solutions over test cases that include Wi-Fi access links. Such evaluations should highlight the inherently different characteristics of Wi-Fi networks in contrast to their wired counterparts:
In summary, the presence of Wi-Fi access links in different network topologies can exert different impact on the network performance in terms of application-layer effective throughput, packet loss rate, and packet delivery delay. These, in turn, will influence the behavior of end-to-end real-time multimedia congestion control.
Unless otherwise mentioned, the test cases in this section choose the PHY- and MAC-layer parameters based on the IEEE 802.11n Standard. Statistics collected from enterprise Wi-Fi networks show that the two dominant physical modes are 802.11n and 802.11ac, accounting for 41% and 58% of connected devices. As Wi-Fi standards evolve over time -- for instance, with the introduction of the emerging Wi-Fi 6 (based on IEEE 802.11ax) products -- the PHY- and MAC-layer test case specifications need to be updated accordingly to reflect such changes.
Typically, a Wi-Fi access network connects to a wired infrastructure. Either the wired or the Wi-Fi segment of the network can be the bottleneck. The following sections describe basic test cases for both scenarios separately. The same set of performance metrics as in [I-D.ietf-rmcat-eval-test]) should be collected for each test case.
We recommend to carry out the test cases as defined in this document using a simulator, such as [NS-2] or [NS-3]. When feasible, it is encouraged to perform testbed-based evaluations using Wi-Fi access points and endpoints running up-to-date IEEE 802.11 protocols, such as 802.11ac and the emerging Wi-Fi 6, so as to verify the viability of the candidate schemes.
The test scenarios below are intended to mimic the setup of video conferencing over Wi-Fi connections from the home. Typically, the Wi-Fi home network is not congested and the bottleneck is present over the wired home access link. Although it is expected that test evaluation results from this section are similar to those as in [I-D.ietf-rmcat-eval-test], it is still worthwhile to run through these tests as sanity checks.
Figure 2 shows the network topology of Wi-Fi test cases. The test contains multiple mobile nodes (MNs) connected to a common Wi-Fi access point (AP) and their corresponding wired clients on fixed nodes (FNs). Each connection carries either a RTP-based media flow or a TCP traffic flow. Directions of the flows can be uplink (i.e., from mobile nodes to fixed nodes), downlink (i.e., from fixed nodes to mobile nodes), or bi-directional. The total number of uplink/downlink/bi-directional flows for RTP-based media traffic and TCP traffic are denoted as N and M, respectively.
Uplink
+----------------->+
+------+ +------+
| MN_1 |)))) /=====| FN_1 |
+------+ )) // +------+
. )) // .
. )) // .
. )) // .
+------+ +----+ +-----+ +------+
| MN_N | ))))))) | | | |========| FN_N |
+------+ | | | | +------+
| AP |=========| FN0 |
+----------+ | | | | +----------+
| MN_tcp_1 | )))) | | | |======| FN_tcp_1 |
+----------+ +----+ +-----+ +----------+
. )) \\ .
. )) \\ .
. )) \\ .
+----------+ )) \\ +----------+
| MN_tcp_M |))) \=====| FN_tcp_M |
+----------+ +----------+
+<-----------------+
Downlink
Figure 2: Network topology for Wi-Fi test cases
The test cases in this section assume that the wired segment along the media path is well-provisioned whereas the bottleneck exists over the Wi-Fi access network. This is to mimic the application scenarios typically encountered by users in an enterprise environment or at a coffee house.
Same as defined in Section 3.1.1
This section describes a few test scenarios that are deemed as important for understanding the behavior of a candidate RTP-based congestion control scheme over a Wi-Fi network.
The EDCA/WMM mechanism defines prioritized QoS for four traffic classes (or Access Categories). RTP-based real-time media flows should achieve better performance in terms of lower delay and fewer packet losses with EDCA/WMM enabled when competing against non-interactive background traffic such as file transfers. When most of the traffic over Wi-Fi is dominated by media, however, turning on WMM may degrade performance since all media flows now attempt to access the wireless transmission medium more aggressively, thereby causing more frequent collisions and collision-induced losses. This is a topic worthy of further investigation.
As discussed in [Heusse2003], the presence of clients operating over slow PHY-layer link rates (e.g., a legacy 802.11b device) connected to a modern network may adversely impact the overall performance of the network. Additional test cases can be devised to evaluate the effect of clients with heterogeneous link rates on the performance of the candidate congestion control algorithm. Such test cases, for instance, can specify that the PHY-layer link rates for all clients span over a wide range (e.g., 2Mbps to 54Mbps) for investigating its effect on the congestion control behavior of the real-time interactive applications.
This memo includes no request to IANA.
The security considerations in [I-D.ietf-rmcat-eval-criteria] and the relevant congestion control algorithms apply. The principles for congestion control are described in [RFC2914], and in particular, any new method must implement safeguards to avoid congestion collapse of the Internet.
Given the difficulty of deterministic wireless testing, it is recommended and expected that the tests described in this document would be done via simulations. However, in the case where these test cases are carried out in a testbed setting, the evaluation should take place in a controlled lab environment. In the testbed, the applications, simulators and network nodes ought to be well-behaved and should not impact the desired results. It is important to take appropriate caution to avoid leaking non-responsive traffic with unproven congestion avoidance behavior onto the open Internet.
The following individuals contributed to the design, implementation, and verification of the proposed test cases during earlier stages of this work. They have helped to validate and substantially improve this specification.
Ingemar Johansson, <ingemar.s.johansson@ericsson.com> of Ericsson AB contributing to the description and validation of cellular test cases during the earlier stage of this draft.
Wei-Tian Tan, <dtan2@cisco.com>, of Cisco Systems designed and set up a Wi-Fi testbed for evaluating parallel video conferencing streams, based upon which proposed Wi-Fi test cases are described. He also recommended additional test cases to consider, such as the impact of EDCA/WMM usage.
Michael A. Ramalho, <mar42@cornell.edu> of AcousticComms Consulting (previously at Cisco Systems) applied learnings from Cisco's internal experimentation to the early versions of the draft. He also worked on validating the proposed test cases in a VM-based lab setting.
The authors would like to thank Tomas Frankkila, Magnus Westerlund, Kristofer Sandlund, Sergio Mena de la Cruz, and Mirja Kuehlewind for their valuable inputs and review comments regarding this draft.
| [Heusse2003] | Heusse, M., Rousseau, F., Berger-Sabbatel, G. and A. Duda, "Performance anomaly of 802.11b", in Proc. 23th Annual Joint Conference of the IEEE Computer and Communications Societies, (INFOCOM'03), March 2003. |
| [HO-def-3GPP] | TR 21.905, 3GPP., "Vocabulary for 3GPP Specifications", December 2009. |
| [HO-LTE-3GPP] | TS 36.331, 3GPP., "E-UTRA- Radio Resource Control (RRC); Protocol specification", December 2011. |
| [HO-UMTS-3GPP] | TS 25.331, 3GPP., "Radio Resource Control (RRC); Protocol specification", December 2011. |
| [NS-2] | "ns-2", December 2014. |
| [NS-3] | "ns-3 Network Simulator" |
| [QoS-3GPP] | TS 23.203, 3GPP., "Policy and charging control architecture", June 2011. |
| [RFC2914] | Floyd, S., "Congestion Control Principles", BCP 41, RFC 2914, DOI 10.17487/RFC2914, September 2000. |