BACKGROUND AND MOTIVATION¶
TL;DR
当前LTE和WIFI的技术设计
| Text Only | |
|---|---|
1 2 | |
问题所在
| Text Only | |
|---|---|
1 2 3 4 5 | |
In this section, we give a brief overview of the LTE network architecture, then expand on the evolution of the integration of WiFi with LTE and motivate the need for an effective traffic management solution for LTE and WiFi networks.
在本章节中，我们首先简要概述LTE网络架构，然后详细阐述WiFi与LTE集成的发展历程，并阐明为LTE和WiFi网络设计有效流量管理解决方案的必要性。
LTE Networks¶
The top-half of Figure 1 shows a simplified 4G LTE network architecture, mainly consisting of two parts: the Evolved Packet Core (EPC) Network and the Radio Access Network (RAN). The EPC or the mobile core network consists of both the control and data plane functions. The control plane functionality is provided by the MME (Mobility Management Entity), HSS (Home Subscriber Server) and the PCRF (Policy and charging rules function). The MME handles session and subscriber management including user authentication, mobility management and idle terminal location management. The HSS includes a database that stores the user profile information while the PCRF manages the service policy and configures the QoS parameters for each user traffic flow. The data plane functionality in the EPC is split between the S-GW (Serving gateway) and the PDN-GW (Packet Data Network gateway). The S-GW acts as a local mobility anchor for user sessions as clients move across base stations. The PDN-GW is connected to multiple S-GWs and routes user traffic towards external networks, while also performing policy enforcement for resource management, packet filtering and charging functions. The RAN includes basestations (or eNodeBs) that perform radio resource management and interference mitigation.
图1的上半部分展示了一个简化的4G LTE网络架构，主要由两部分组成：演进分组核心（EPC）网络和无线接入网（RAN）。EPC即移动核心网，包含控制平面和数据平面功能。控制平面功能由MME（移动性管理实体）、HSS（归属用户服务器）和PCRF（策略与计费规则功能）提供。MME负责会话和签约用户管理，包括用户认证、移动性管理和空闲终端位置管理。HSS包含一个存储用户配置信息的数据库，而PCRF管理服务策略并为每个用户数据流配置QoS参数。EPC中的数据平面功能由S-GW（服务网关）和PDN-GW（分组数据网络网关）分担。S-GW在客户端跨基站移动时充当用户会话的本地移动性锚点。PDN-GW连接到多个S-GW，并将用户流量路由至外部网络，同时执行资源管理、数据包过滤和计费功能的策略。RAN包括基站（或eNodeB），它们执行无线资源管理和干扰抑制。
WiFi Integration in Operator Networks¶
Several standard bodies such as 3GPP and WiFi Alliance (WFA) have defined solutions for the network integration of LTE and WiFi. These solutions are mainly classified into two types: (i) Access control: To enable subscriber validation, seamless authentication and billing across LTE and WiFi networks is an important step to ensure WiFi integration. However, the method of authentication varies across operators. Most of the operators provide SIM-based authentication [14] enabling them to maintain a unified subscriber database for both their LTE and WiFi networks, while other operators have adopted the traditional web-based authentication which requires the users to enter their credentials in the browser. (ii) Dataplane integration: To enable offloading capabilities and seamless mobility between LTE and WiFi networks, a tight data-plane integration is required across the networks. Such integration involves the backhauling of WiFi traffic to the LTE core network. Specifically, 3GPP has standardized the I-WLAN architecture [15] to integrate WiFi traffic into LTE’s mobile core network. The architecture as shown in Figure 1 enables the integration using the ePDG (evolved Packet Data Gateway), which serves as a gateway connecting the WiFi access points with the PDN gateway. IPsec tunnels are established between each mobile device and the ePDG, and the IP address is anchored at the PDN gateway. Since the IP address is maintained across the WiFi and LTE networks, flows can be seamlessly migrated across the networks. The PMIPv6 protocol is employed and the ePDG updates the IP address binding at the PDN gateway after authentication and tunnel establishment with the mobile device. Although tight integration will enable operators to ensure policy control, better QoE management and seamless mobility over their networks, it has been resisted by most operators due to the significant increase in backhauling costs.
诸如3GPP和WiFi联盟（WFA）等多个标准组织已经为LTE和WiFi的网络集成定义了解决方案。这些解决方案主要分为两类：(i) 接入控制：实现签约用户验证、跨LTE和WiFi网络的无缝认证和计费是确保WiFi集成的重要步骤。然而，不同运营商的认证方法各异。多数运营商提供基于SIM卡的认证[14]，使其能够为LTE和WiFi网络维护统一的用户数据库，而其他运营商则采用传统的基于Web的认证，需要用户在浏览器中输入其凭证。(ii) 数据平面集成：为实现LTE和WiFi网络间的流量卸载能力和无缝移动性，网络间需要紧密的数据平面集成。这种集成涉及将WiFi流量回传至LTE核心网。具体而言，3GPP已标准化I-WLAN架构[15]，以将WiFi流量集成到LTE的移动核心网中。如图1所示的该架构通过ePDG（演进分组数据网关）实现集成，ePDG作为连接WiFi接入点与PDN网关的网关。在每个移动设备和ePDG之间建立IPsec隧道，并且IP地址锚定在PDN网关。由于IP地址在WiFi和LTE网络间得以保持，数据流可以在网络间无缝迁移。采用PMIPv6协议，并且ePDG在与移动设备进行认证和隧道建立后，在PDN网关更新IP地址绑定。尽管紧密集成将使运营商能够确保策略控制、更好的QoE管理及其网络上的无缝移动性，但由于回传成本显著增加，大多数运营商对此表示抵制。
Current Deployments¶
In the near future, it is expected that operators will transition to using their WiFi networks for new services and revenue generation and provide better QoE for their users rather than just offloading for coverage or during congestion. Moreover, operators are quickly upgrading their network to LTE that offer superior rates than 3G networks and are deploying WiFi APs in areas of high network access. However, current deployments are not designed to use the LTE and WiFi network optimally to ensure good QoE for applications and users. Although most devices are pre-configured with connection managers, they mainly implement functions for network discovery, selection and authentication.
Short-comings: We bring to light a few key issues with current deployments through experiments on our LTE testbed and address them in the design of ATOM. The experiments are conducted using a network of a single LTE basestation and a WiFi AP.
预计在不久的将来，运营商将过渡到使用其WiFi网络提供新服务和创造收入，并为其用户提供更好的QoE，而不仅仅是在覆盖或拥塞期间进行流量卸载。此外，运营商正在迅速将其网络升级到比3G网络速率更优的LTE，并在网络接入密集的区域部署WiFi AP。然而，当前的部署并非旨在优化使用LTE和WiFi网络以确保应用和用户的良好QoE。尽管大多数设备都预配置了连接管理器，但它们主要实现网络发现、选择和认证等功能。
不足之处 (Short-comings)： 我们通过在我们的LTE测试平台上进行的实验，揭示了当前部署中存在的一些关键问题，并在ATOM的设计中着手解决这些问题。实验使用一个由单个LTE基站和单个WiFi AP组成的网络进行。
(i) Naive policies: Most connection managers [5] are configured with simple policies that ensure the device connects to a WiFi AP in case a connection is made. A few connection managers use the WiFi interface only if the signal strength is above some threshold. However since they do not take the current load on the AP into account, the QoE of the users could suffer during congestion. To drive our point, we setup an experiment such that 6 users are randomly distributed and are within the coverage of the WiFi AP, while 2 users are outside the coverage of the WiFi AP. All the 8 users stream videos from YouTube with an average bit-rate of about 2 Mbps. We plot the throughput obtained by 3 out of the 6 WiFi users and the 2 LTE users in Figures 2(a) and (b) respectively. We see that the throughput of WiFi users is less than the average bit-rate (2Mbps) of the video resulting in stalls in the video stream while the throughput of LTE users is above the average bit-rate resulting in a smooth stream. Figure 2(c) depicts the resource utilization: while the WiFi AP is over-utilized, the utilization of the LTE basestation is only 25%.
(i) 简单策略 (Naive policies)： 大多数连接管理器[5]配置了简单的策略，确保设备在建立连接时连接到WiFi AP。少数连接管理器仅在信号强度高于某个阈值时才使用WiFi接口。然而，由于它们没有考虑AP上的当前负载，在拥塞期间用户的QoE可能会受到影响。为证明我们的观点，我们设置了一个实验：6名用户随机分布在WiFi AP的覆盖范围内，另2名用户在WiFi AP的覆盖范围之外。所有8名用户都从YouTube流式传输视频，平均比特率约为2 Mbps。我们在图2(a)和(b)中分别绘制了6名WiFi用户中的3名以及2名LTE用户的吞吐量。我们看到，WiFi用户的吞吐量低于视频的平均比特率（2Mbps），导致视频流出现卡顿，而LTE用户的吞吐量高于平均比特率，从而实现了流畅的流媒体播放。图2(c)描述了资源利用率：WiFi AP被过度利用，而LTE基站的利用率仅为25%。
(ii) Static decisions: Moreover, it is not sufficient to make interface selection decision at the initiation of a user flow as wireless conditions change significantly due to user arrival/departure and mobility. To drive our point, we use a similar setup with 4 users on the WiFi AP. As shown in Figure 2(d), initially all the WiFi users receive throughput in excess of the video bit-rate. At around 10 seconds, we move a couple of the WiFi users away from the AP at walking speeds. As a result of the user mobility, the WiFi AP is unable to support the video rates of its users as shown in Figure 2(d) affecting the video of the users mid-stream. However, to enable dynamic traffic management, operators are required to poses the capability to switch the interface of user flows seamlessly across their LTE and WiFi networks. Such a capability needs tight dataplace integration of the WiFi network with the LTE network. While the integration of access control (authentication) methods for WiFi have been widely adopted by operators [4], tighter integration of data or bearer plane to the LTE network has been resisted by most operators, mainly due to: (1) Backhauling large amounts of WiFi traffic through their LTE core network significantly increases both Operational costs (OP-EX) in terms of backhaul costs and Capital costs (CAP-EX) in order to scale their LTE core gateways. (2) Most of the traffic and services on mobile networks is OTT (Over-the-top) that does not generate direct revenue for the operators. Hence there is little incentive for operators to invest significantly in order to provide QoE for such services. (3) In most scenarios, we discovered that the WiFi business units of operators are managed independently from the LTE business.
(ii) 静态决策 (Static decisions)： 此外，由于用户到达/离开和移动性导致无线条件显著变化，仅在用户流发起时做出接口选择决策是不够的。为证明我们的观点，我们使用了类似的设置，其中4名用户位于WiFi AP上。如图2(d)所示，最初所有WiFi用户接收到的吞吐量都超过了视频比特率。大约在10秒时，我们让几名WiFi用户以步行速度离开AP。由于用户的移动性，WiFi AP无法支持其用户的视频速率，如图2(d)所示，影响了用户正在播放的视频流。然而，要实现动态流量管理，运营商需要具备在其LTE和WiFi网络间无缝切换用户流接口的能力。这种能力需要WiFi网络与LTE网络进行紧密的数据平面集成。虽然WiFi的接入控制（认证）方法的集成已被运营商广泛采用[4]，但大多数运营商对与LTE网络进行更紧密的数据或承载平面集成持抵制态度，主要原因如下：(1) 通过其LTE核心网回传大量WiFi流量，会在回传成本方面显著增加运营成本（OP-EX），同时为了扩展其LTE核心网关也会增加资本支出（CAP-EX）。(2) 移动网络上的大部分流量和服务是OTT（Over-the-top）业务，这些业务不为运营商产生直接收入。因此，运营商缺乏为提供此类服务的QoE而进行重大投资的动力。(3) 在大多数情况下，我们发现运营商的WiFi业务部门与LTE业务部门是独立管理的。
(iii) Coarse-grained policies[6, 7]: Operators will desire the ability to perform interface selection on a per-application level rather than a per-user or per-device level. This capability ensures (a) operators can provide QoE depending upon the application requirements and (b) content providers may be willing to pay mobile operators for better QoE for users accessing their applications in the future. Operators will need to differentiate the performance of such flows over other OTT traffic. We conduct an experiment to show the disadvantage of the inability to perform fine-grained traffic management. The experiment is setup with 8 LTE users within the coverage of the WiFi AP and 4 LTE users outside the WiFi coverage. All the 8 users download a large file from the WiFi AP. One of the WiFi users (User#5) also streams a YouTube video of average rate 2Mbps. All the 4 LTE users stream the same YouTube video from the LTE basestation. Figure 2(e) plots the average number of stalls in the video session of the 4 LTE users and User#5. Scenario 1 represents the case where all the traffic of User#5 is mapped to the WiFi AP since the user is within the coverage of the AP. Clearly, the video flow of User#5 suffers significantly as the WiFi AP is congested. Scenario 2 represents the case with user-level traffic management where both the flows of User#5 (video and file-download) are moved to the LTE network. This results in the LTE network getting congested and the video of all the 5 users suffer. A fine-grained traffic management solution would move the video flow of User#5 to LTE while keeping the file-download flow on the WiFi AP, resulting in good performance for the video of all the 5 users.
(iii) 粗粒度策略 (Coarse-grained policies)[6, 7]： 运营商希望能够实现基于单个应用的接口选择能力，而不仅仅是基于单个用户或单个设备。这种能力确保 (a) 运营商可以根据应用需求提供QoE，以及 (b) 内容提供商未来可能愿意为访问其应用的用户获得更好的QoE而向移动运营商付费。运营商将需要区分此类流量与其他OTT流量的性能。我们进行了一项实验，以显示无法执行细粒度流量管理的缺点。实验设置为8名LTE用户在WiFi AP覆盖范围内，4名LTE用户在WiFi覆盖范围外。所有8名在WiFi覆盖范围内的用户都从WiFi AP下载一个大文件。其中一名WiFi用户（用户#5）同时流式传输一个平均速率为2Mbps的YouTube视频。所有4名在WiFi覆盖范围外的LTE用户都从LTE基站流式传输相同的YouTube视频。图2(e)绘制了4名LTE用户和用户#5视频会话中的平均卡顿次数。场景1表示用户#5的所有流量都映射到WiFi AP的情况，因为该用户在AP的覆盖范围内。显然，由于WiFi AP拥塞，用户#5的视频流受到严重影响。场景2表示采用用户级流量管理的情况，用户#5的两个流（视频和文件下载）都被移至LTE网络。这导致LTE网络拥塞，所有5名用户的视频都受到影响。一个细粒度的流量管理解决方案会将用户#5的视频流移至LTE，同时将其文件下载流保留在WiFi AP上，从而为所有5名用户的视频带来良好性能。