RELATED WORK

International roaming has received little coverage in terms of large measurement studies, potentially because of the high costs and coordination efforts associated with running such a campaign. Vallina et al. [26] has leveraged crowdsourced measurements and focused only on national roaming agreements between MNOs in France. The study does not provide any further evaluation in terms of performance or content implications. Using controlled measurements in the dedicated platform MONROE [8] enabled Michelinakis et al. [22] to analyze the impact of international roaming, but only for two operators in Europe. They find that the home-routed configuration does impact the performance of cloud service providers, such as Akamai or CloudFront. Our paper complements this work and presents an extensive measurement study to understand the international roaming ecosystem in Europe since the “Roam like Home” initiative.

就大规模测量研究而言，国际漫游领域获得的关注甚少，这可能是因为开展此类活动所需的高昂成本和大量的协调工作。Vallina等人[26]利用了众包测量数据，并且仅关注法国移动网络运营商（MNO）之间的国内漫游协议。该研究未在性能或内容影响方面提供任何进一步的评估。Michelinakis等人[22]利用专用平台MONROE[8]中的受控测量，分析了国际漫游的影响，但其研究仅针对欧洲的两家运营商。他们发现，归属路由（home-routed）配置确实会影响如Akamai或CloudFront等云服务提供商的性能。本文是对这项工作的补充，它展示了一项广泛的测量研究，旨在深入理解自“漫游如在家”（Roam like Home）倡议实施以来欧洲的国际漫游生态系统。

There have been myriad recent studies focusing on mobile network characterization and performance. For instance, while Huang et al. [15] study LTE network characteristics in a cellular operator in the US, Safari et al. [18] show performance measurement in mobile networks are much more complex than wired networks, due to the different network configurations such as the presence of Network Address Translation (NAT) or Performance Enhancing Proxies (PEP), which do vary over time. Kaup et al. [17] run a crowdsourcing campaign to measure RTT and throughput towards popular websites in Germany. They used the dataset to show that the association of a mobile endpoint to the Point of Presence (PoP) within the operator network has influence on network performance. The authors of [23] present a mobile app and a mechanism for identifying traffic differentiation for arbitrary applications in the mobile networks. Ververis et al. [27] surveys content filtering for a mixture of broadband and cellular ISPs and finds a lack of transparency around the policies they implement as well as outdated and poorly implemented blacklists. In our work, we not only focus on network performance of roaming infrastructure, but also identify possible traffic differentiation for particular applications and content discrimination and geo-restriction for users in international roaming.

近期有大量研究关注移动网络的特征分析和性能。例如，Huang等人[15]研究了美国一家蜂窝网络运营商的LTE网络特性；而Safari等人[18]指出，由于存在随时间变化的网络地址转换（NAT）或性能增强代理（PEP）等不同网络配置，移动网络中的性能测量远比有线网络复杂。Kaup等人[17]在德国开展了一项众包活动，以测量指向热门网站的RTT（往返时间）和吞吐量。他们利用该数据集表明，移动端点与运营商网络内接入点（PoP）的关联对网络性能有影响。[23]的作者们提出了一款移动应用程序和一种机制，用于识别移动网络中针对任意应用的流量差异化。Ververis等人[27]调查了混合宽带和蜂窝ISP（互联网服务提供商）的内容过滤情况，发现他们实施的策略缺乏透明度，并且使用了过时且执行不佳的黑名单。在我们的工作中，我们不仅关注漫游基础设施的网络性能，还识别了针对特定应用的可能存在的流量差异化，以及针对国际漫游用户的内容歧视和地理限制问题。