跳转至

DATA ROAMING TRAFFIC

In this section, we explore the data roaming dataset (see Table 1), and give further insights into the type of traffic flowing through the IPX-P’s platform and its performance.

Tip
  1. 数据漫游流量中，绝大多数为TCP与UDP协议，分别主要承载网页访问（HTTP/HTTPS）与DNS（端口53）流量
  2. 在性能方面，RTT 主要受到漫游配置模式（归属路由或本地中断）以及地理位置的影响
  3. 连接建立时延 与 会话持续时间 则更依赖于具体应用场景和所连接的远程服务器

Roaming Traffic Breakdown

Each data record captures different applications/protocols that correspond to the subscriber activity over the period we monitor. The data record integrates performance parameters per roaming data communication, including Round-Trip Time (RTT), packet retransmissions or volume of bytes transferred (uplink and downlink). We find that the majority of the traffic we capture is TCP (40%) or UDP(57%), with a small fraction of ICMP (2%) and other protocols. The destination port breakdown for the TCP records we capture reveals that the largest amount of traffic is web traffic (HTTP, HTTPS), accounting for 60% of TCP traffic. The breakdown of UDP traffic shows that majority of traffic we capture is DNS over port 53 traffic (more than 70%). This is (largely) due to the procedure the MNOs roaming partners implement in order to allow for international context creation (tunnels) over the IPX Network (i.e., control traffic). The Visited Mobile Network Operator (VMNO) uses the IPX to resolve the Access Point Name (APN) associated to the mobile subscriber to an actual IP address corresponding to the home network GGSN (or PGw for EPC), which the Home Mobile Network Operator (HMNO) performs.

每条数据记录涵盖了不同的应用与协议，反映了被监测期间用户的活动行为。每条记录整合了与数据漫游通信相关的性能指标，包括往返时延（RTT）、分组重传情况以及传输的数据体量（上行与下行字节数）。

我们发现所采集的 大部分流量为TCP（约占40%）或UDP（约占57%），另有少量为ICMP（约2%）及其他协议

从 TCP流量 的目标端口分析可见，其中 60%的流量为网页访问（HTTP与HTTPS）

UDP流量 的分析结果显示， 绝大多数流量为通过53号端口的DNS流量 （占比超过70%）

这在很大程度上源自MNO漫游合作方为支持通过IPX网络建立国际数据上下文（即隧道）所采用的流程，即这类流量多为控制类信令。漫游移动网络运营商（VMNO）使用IPX解析与移动用户相关的接入点名称（APN），将其映射至归属网络的GGSN（或EPC架构下的PGw）的实际IP地址，该解析操作最终由归属移动网络运营商（HMNO）完成。

Performance Implications

In this section, we investigate the quality of the roaming service over the IPX-P’s network. We focus our analysis on devices operating with IMSIs from a Spanish operator that supports multiple IoT verticals (e.g., energy sensors, fleet tracking, wearables, etc.) over the world. We zoom into the top countries in terms of number of devices (namely, UK, Mexico, Peru, US and Germany), and monitor the session duration, RTT uplink and downlink, and the connection setup delay for all the TCP data communications the IPX-P supports during the period of analysis (July 2020). The session duration (Figure 13a) varies largely, depending on the country where these IoT devices roam and, likely, the usage dictated by the IoT provider deploying these devices. We note that the devices in Germany have significantly longer average session duration (≈45s) than devices in the UK (≈150s).

本节中，我们进一步分析IPX-P网络中数据漫游服务的质量表现。我们聚焦于使用西班牙运营商IMSI号段的设备，这些设备被部署于多个物联网垂直场景（例如能源传感器、车队追踪、可穿戴设备等），广泛分布于全球。我们重点分析漫游设备数量最多的国家（即英国、墨西哥、秘鲁、美国和德国），并监测在2020年7月分析期间内，IPX-P所支持的TCP数据通信的会话持续时间、上下行RTT以及连接建立时延等性能指标。

如图13a所示，会话持续时间因设备所在国家及部署用途的不同而显著差异。我们注意到，在德国的设备平均会话时长约为45秒，明显短于在英国的设备（约150秒）。这种差异很可能与设备用途或部署策略相关。

For the TCP traffic, we further check the RTT distribution the mobile subscribers experiment, broken down per visited country. The uplink RTT (Figure 13b) captures the RTT between the sampling point (i.e., Miami) and the application server, capturing the impact of the PGw (or the GGSN) and the latency over the Internet path towards the application server. The downlink RTT (Figure 13c) captures the RTT between the sampling point within the IPX-P’s infrastructure and the mobile subscriber, thus capturing the impact of the visited network (including the radio access network) and the SGw (or the SGSN, respectively). For both metrics, we notice that the lowest values are for devices operating in the US. This is due to the use of a different roaming configuration, namely local breakout roaming configuration, in this visited MNO. We note that, in the case of home routed roaming configuration, the uplink RTT might vary with the distance between the home country (in this case, Spain) and the visited country. This is reflected in the distance between the PGw (within the Spanish operator’s network) and the application server (likely within the visited country). The downlink RTT shows similar pattern and rank between the groups of devices per visited country.

对于TCP流量，我们进一步考察了不同国家用户所体验的 RTT分布情况

上行RTT（图13b）表示从采样点（位于迈阿密）到应用服务器的往返时延，反映了PGw（或GGSN）及IP网络路径的延迟影响；而下行RTT（图13c）表示从IPX-P基础设施内采样点到用户设备之间的往返时延，反映了访问网络（包括无线接入网）和SGw（或SGSN）的性能表现。对于这两个指标，我们发现设备在美国的RTT最小，这得益于其采用了不同的漫游配置方式，即本地中断（Local Breakout）模式。

相对而言，若采用归属路由（Home Routed）漫游配置，上行RTT则会因归属国（此处为西班牙）与漫游国之间的距离而有所增加，这也体现在PGw（位于西班牙网络中）与目标应用服务器（通常位于漫游国）的网络距离上。下行RTT的趋势与上行类似，在不同国家的设备群体之间表现出一致的排序规律。

The connection setup delay (Figure 13d) represents the time in milliseconds between TCP SYN (first packet sent by the mobile subscriber) and TCP ACK (last packet in the three handshake procedure). We observe that this metric does not follow the same trends of the RTTs. This highlights the applications/IoT verticals and of remote servers play a dominant role in the connection setup delay.

TCP连接建立时延 （图13d）是指移动用户发送第一个TCP SYN包到完成三次握手中的最后一个ACK包之间的时间（单位为毫秒）。我们发现，该指标并未表现出与RTT完全一致的趋势，说明应用场景、物联网垂直领域及所连接的远程服务器在连接建立过程中的角色更加显著。

Takeaway: Majority of the data roaming traffic is TCP or UDP used for Web (i.e., HTTP/HTTPS) and DNS over port 53, respectively. In terms of performance, the RTTs strongly depend on the roaming configuration (i.e., home routed or local breakout), and is then impacted by the geographical location of the users, while the connection establishment delay and session duration is dominated by the applications/IoT verticals and remote servers. The IPX-P leverages the flexibility of the IPX model to offer tailored connectivity to different customers, depending on their requirements.

总结:

  1. 数据漫游流量中，绝大多数为TCP与UDP协议，分别主要承载网页访问（HTTP/HTTPS）与DNS（端口53）流量
  2. 在性能方面，RTT 主要受到漫游配置模式（归属路由或本地中断）以及地理位置的影响
  3. 连接建立时延 与 会话持续时间 则更依赖于具体应用场景和所连接的远程服务器

IPX-P平台灵活地利用IPX网络模型，为不同客户提供差异化的连接服务，以满足其多样化的需求