SS7/DIAMETER SIGNALING

In this section, we provide insights into the complexity of the operations of the IPX-P by analyzing one of the base functions it provides, namely, signaling between mobile core network elements – SCCP Signaling and Diameter Signaling. These functions are mandatory for the correct operation of data roaming and multiple other IPX-P services. The SCCP Signaling and Diameter Signaling datasets we collect from the IPX-P comprise signaling information for 2G/3G and 4G/LTE radio technologies from all the devices that use the IPX-P’s infrastructure (Table 1).

在本节中，我们通过分析IPX-P提供的一项基础功能，即 移动核心网元之间的信令交互——SCCP信令与Diameter信令 ，来洞察其运营的复杂性。这些功能对于数据漫游及IPX-P多种其他服务的正确运行是必不可少的。我们从该IPX-P收集的SCCP信令与Diameter信令数据集，包含了所有使用该IPX-P基础设施的设备所产生的2G/3G和4G/LTE无线技术的信令信息（见表1）。

Signaling Traffic Trends

Figure 3 shows signaling activity of roaming mobile subscribers during the observation period in July 2020. We look at both Signaling System No. 7 (SS7) and Diameter signaling procedures. MAP is the most important application protocol in the SS7 stack, and handles the roamers’ mobility between countries in 2G-3G. The same function is performed by the Diameter [13] signaling protocol in LTE.

Overall, we capture more than 120M devices active in the MAP dataset, and more than 14M devices active in the Diameter dataset. 3 This shows a slight decrease compared with December 2019 (likely due to mobility restrictions imposed to tackle the COVID-19 emergency [20]), when the total number of IMSIs we captured was more than 130M active in 2G/3G and more than 15M active in 4G/LTE. These results highlight that the IPX-P 2G/3G infrastructure handles an order of magnitude more devices than the 4G infrastructure.

Figure 3a shows the average number of records per IMSI calculated over all the IMSIs we observe in each one-hour interval (continuous line) during the July 2020 observation period, as well as the standard deviation of the number of records per IMSI calculated over all the IMSIs active in the same one hour interval (shaded area). We observe both the MAP procedures for 2G/3G (red color) and the similar Diameter procedures for 4G/LTE (green color). Each record in both of these datasets represents a signaling dialogue that two network elements have, corresponding to different standard procedures. For instance, from the MAP interface we capture mobility management routines, including location management and authentication. While Diameter and MAP are different protocols, the underlying functional requirements (e.g., authenticating the user to set up a data communication) have many similarities in terms of the messages used for Diameter and the SS7 MAP protocol implementation. We note that the load in terms of average signaling records per IMSI is in the same order of magnitude (the continuous lines on the plot), regardless of the infrastructure the devices use; yet, there are significantly more messages generated on average by an IMSI using MAP than an IMSI using Diameter, as Diameter is a more efficient protocol than MAP [13, 30].

We further break down the signaling traffic on record type (or procedure) both for MAP (Figure 3b and Diameter (Figure 3c). Figure 3b shows the time series of signaling traffic broken down by type of procedure, including Update Location (UL), Cancel Location (CL) and Send Authentication Information (SAI) messages. The latter, SAI, represents the highest fraction of MAP signaling traffic; this is also the case for the Diameter signaling traffic. Indeed, according to the GSM standard definition, the visited network triggers the authentication of subscriber procedure upon IMSI attach, location update or before starting data communication, thus explaining the larger volume of SAI messages.

Takeaway: We find that the number of devices using the IPX-P’s 2G/3G infrastructure (MAP traffic) is an order of magnitude higher than those using 4G infrastructure (Diameter traffic). The volume of signaling traffic in the SCCP infrastructure is, correspondingly, more significant than in the Diameter infrastructure. This heavy reliance on 2G/3G is problematic, because of the high costs the maintaining legacy radio networks incurs to operators. This brings to light the lack of global consistency between operators in deploying the latest generation access technologies. Further, the use of less efficient protocols imposes a higher operational cost for both the IPX-P platform and its customers.

Operational Breadth

The goal the IPX-P is to offer global coverage to its entire customer base. This means allowing all customer devices to connect anywhere in the world, and, conversely, allowing anyone in the world to connect to their customers’ networks. Overall, the IPX-P’s infrastructure serves devices from MNOs from over 220 (home) countries, operating in more than 210 (visited) countries. In Figure 4 we show the distribution of mobile devices, and focus on top-14 home operators and top-14 visited operators in July 2020. We notice that the distribution is fairly skewed to few operators, and the best represented countries correspond to the locations of the main IPX-P’s customers, namely Spain, UK, Germany.

Through the lens of the signaling dataset, we can further capture the (international) mobility of devices. Figure 5 shows the distribution of mobile devices, registered during each of the two observation intervals (December 2019 in Figure 5a and July 2020 in Figure 5b), that travel from their home country (column) to a visited country (row). In the following we comment on the December 2019 dataset. Overall, we find that the majority of subscribers using the IPX-P infrastructure - serving large European MNOs – comes from UK (≈8 million devices in December 2019), Germany (≈2 million devices) or Spain (≈2 million devices). Most of these devices tend to visit the UK (≈6.5 million devices), Germany (≈2.5 million devices) or the US (≈500,000 devices).

When clustered by geographic regions (i.e, Europe and the Americas), we see that the most popular destinations roamers visit include the UK in Europe and the US in America. Indeed, Figure 5a shows that the UK operators connected to the IPX-P we analyze receive 34% of all the devices from Germany (DE) visible in the system, 85% of the devices from the Netherlands (NL), and 45% of all devices from Spain (ES), among others. Interestingly, we verified with the British operator connected to our IPX-P and found that the inbound roaming devices from the Netherlands (≈ 7.8 million devices) are IoT devices deployed by energy providers (smart meters) 4 . US, Brazil and Mexico emerge as the most popular destinations in the Americas. Specifically, the US operators connected to our IPX-P accommodate 79% of all the outbound roaming devices from Mexico (MX) using the IPX-P infrastructure, 44% of all outbound roaming devices from El Salvador (SV), 17% of all outbound roaming devices from Colombia (CO) and 22% of all outbound roaming devices from Brazil (BR).

Finally, it is interesting to note how data from an IPX-P can capture socio-economic patterns in international mobility. Indeed, we can observe the migration between Venezuela and Colombia, with 71% of the subscribes from Venezuela (VE) traveling to Colombia (CO) during the period we capture. Inversely, we find that 56% of all Colombian outbound roamers travel to Venezuela (VE). The Venezuela-Colombia border is one of the most active in the world [11], as Colombia is the primary destination of most Venezuelan migrants, which capture Venezuelans with different status ranging from economic migrants to refugees.

We observe in both the observation periods a fraction of devices that operate within their home countries. For instance, in July 2020 (Figure 5b), we note that 39% of all the UK devices operate within their home country, or 47% of Mexico device operate within Mexico. These usually belong to Mobile Virtual Network Operatorss (MVNOs) enabled by the IPX-P we analyze, to operate on top of MNOs that are already customers of the MNO. The increased ratio in July 2020 is a side-effect of reduced international mobility in July 2020, compared to December 2019.

Takeaway: The IPX-P underlying infrastructure impacts its operational breadth. Specifically, given that the IPX-P leverages access to important trans-oceanic infrastructure connecting the Americas and Europe (e.g., Brusa subsea cable connecting Brazil and USA, Marea subsea cable connecting the US and Spain, or the SAm-1 subsea cable with various landing points from US to Argentina), we note that these are the main markets where it operates. Specifically, US, UK, MX and BR emerge as the main mobility hubs for devices that depend on this particular IPX-P to operate and infrastructure needs to be provisioned accordingly. At the same time, IPX-P operations need to provide coverage beyond this core infrastructure to more than 200 countries, and can be impacted by specific socio-economic trends.

Steering of Roaming

Every dialogue we capture in our signaling dataset corresponds to a roaming procedure and includes, apart from the requested operation code, the result of the operation. For example, in the case of roamer authentication in the visited network, the requested operation code is "Send Authentication Information", to which the home network replies with the requested information. This is the most frequent procedure we capture in our dataset (Figure 3). In the event of a failure, the response from the home network may contain an error code showing why the procedure failed. Such errors for the SAI operation include Unknown Subscriber (There is no allocated IMSI or no directory number for the mobile subscriber in the home network), and for the UL operation include Unexpected data value (The data type is formally correct, but its value or presence is unexpected in the current context.).

Figure 6 shows the time-series of errors in the MAP dataset, regardless of the type of the operation that triggered them, broken down per type of error, for the July 2020 dataset. We note that the most frequent error is Unknown Subscriber, pointing to a numbering issue during the SAI procedure.

Another frequent error code we observe is Roaming not Allowed (i.e., the home operator is baring the roaming of the device), which corresponds to an Update Location procedure. Often, operators use this error code to implement different routing policies for the mobile user, such as Steering of Roaming (SoR) [6]. In a general manner, this may bring an increase of the signaling load between 10% and 20% [6]. By using SoR, an MNO can specify the preferred roaming partner in a given visited country and allow the IPX-P to enforce those preferences. With this in place, if a roamer device traveling outside its home network (HMNO) 5 attempts to attach to a less preferred roaming partner, the IPX-P will force the Roaming Not Allowed response code (RNA, error code = 8) to an Update Location (UL) message intercepted from the visited network (VMNO). The IPX-P will then try steering the roamer to one of the HMNO’s preferred roaming partners after forcing four UL attempts from the roamer to fail, unless no preferred roaming partner is available in the area (in which case, the SoR platform triggers an exit control to avoid the risk of the roamer not receiving service at all).

Figure 7 shows the percentage of end-user devices roaming from the home country (column-wise) to a visited country (row-wise) for which we registered at least one RNA error code for the UL procedure. We observe a non-negligible number of Roaming Not Allowed operational code as a result of the UL request from the VMNO, typically due to the use of the SoR service the IPX-P provides its customers.

One notable exception is Venezuela. We note the prevalence of this error code for mobile subscribers traveling from Venezuela abroad, regardless the visited country. Because of the volatility of Venezuelan currency, mobile operators in Venezuela suspended international roaming as they said they lacked enough foreign currency to pay roaming partners in foreign countries. The reason this is allowed for Spain (where we only note that 20% of subscribers from Venezuela receive a RNA message) is because of internal agreements between operators that belong to the same international corporation.

On the other end of the spectrum, we see that the fraction of UK users (marked GB in Fig. 7) affected by this error code is very small, regardless of the country they visit. This is because the IPX-P’s customer in the UK does not use the SoR service from the IPX-P, but instead handles the steering of its subscribers separately. Thus, the RNA errors we capture are due to the HMNO from UK not allowing its subscribers to roam (e.g., because of billing issues).

Takeaway: Operators use forced errors to implement different policies for their subscribers when these are roaming abroad. One example is the Steering of Roaming, which the IPX-P offers as a service for its customers, at the cost of increasing the signaling load on the roaming platform.

Impact of IoT Devices

In the following paragraphs we focus on the traffic corresponding to the IoT devices that the IPX-P M2M platform operates for the December 2019 dataset. Although not reported, analysis of the July 2020 dataset leads to similar takeaways.

Figure 8 shows the time series of average number of signaling messages per device for 2G/3G (Fig. 8a) and 4G/LTE (Fig. 8b), as well as the 95th percentile calculated over one hour intervals. To put this figures in context, we include statistics from a similar number of smartphones using the same radio technology. We selected the set of smartphones leveraging the device brand information, which we retrieve by checking the International Mobile Equipment Identity (IMEI) and the corresponding Type Allocation Code (TAC) code, and included only iPhone and Samsung Galaxy devices (the two most popular smartphones) in the pool. Figure 8 shows that IoT devices generally trigger a higher load on the signaling infrastructure, regardless of the infrastructure they use (Diameter or SS7). This holds when checking either the average number of messages per device across time as well as the 95% percentile per hour across all devices.

We also compare the duration of roaming sessions (i.e., the total number of days a device sent at least one signaling message while in roaming) for both IoT devices and smart phones. Figure 9a shows the number of days an IoT device was active during the twoweek period we analyze. We note that the majority of IoT devices have long roaming sessions, which in our case cover the entire observation period. This is very different to what we observe for smartphones (Figure 9b), whose roaming session lengths are shorter. This is expected, since IoT devices are meant to provide services in the country where the IoT provider deploys its services during long periods of time, thus becoming "permanent roamers" in the visited country. At the same time, this also translates into significant signaling load on the VMNO infrastructure from inbound roaming IoT devices.

Takeaway: The M2M service of the IPX-P is very popular and has different operational requirements than the other services. IoT devices operate as permanent roamers with long roaming session, and generate more signaling traffic than smartphones. This contributes significantly more traffic to the IPX-P signaling system than smartphone devices.