The eSIM Stress Test: When Connectivity Is Under Pressure
A travel eSIM can deliver impressive speed and still struggle through a working day. Alertify examines what happened when video calls, VPN connections, cloud synchronization and live streaming competed for the same mobile connection across four international destinations.
The connection is fast. The signal is strong. Your speed test looks excellent. Then the client meeting begins, your laptop starts synchronizing files, and the VPN connects. Nothing has disconnected, but suddenly everything feels slower.
This is one of the most revealing problems in modern mobile connectivity.
Network performance is usually presented as a number: download speed, upload speed, or sometimes latency.
Those figures tell us something about a connection at a particular moment. They don't necessarily tell us how that connection behaves when several applications need it simultaneously.
A business traveler might join a Zoom meeting while a corporate VPN is active. A cloud storage service may synchronize documents in the background. Messaging applications may continue refreshing. A browser may download large files without the traveler consciously initiating every transfer.
Individually, these activities are ordinary. Together, they create a demanding workload for a mobile connection.
That is the distinction investigated in Alertify Research #03: The eSIM Stress Test.
Our first study examined latency, jitter and packet loss. The second investigated the network architecture that can influence those measurements.
The third introduces a different variable: simultaneous demand.
We revisit the original observations from Medellín, Chiang Mai, Istanbul and Athens to understand why some connections remained stable during demanding workloads while others became progressively less usable.
The central finding goes beyond the differences between individual eSIM plans.
A mobile connection does not have to disconnect to fail the user. It can remain connected while losing the responsiveness and consistency required to complete important work.
Three findings that conventional speed tests rarely reveal.
A good idle connection can deteriorate quickly under load.
In Medellín, some connections that initially recorded latency below 70 ms later exceeded 180 ms during combined Zoom and VPN activity. Idle performance did not predict loaded responsiveness.
Background synchronization can disrupt the foreground task.
During the Istanbul observations, transferring a roughly 500 MB cloud folder while sharing a screen exposed interruptions that were less apparent during lighter usage.
A provider's regional and global products can behave differently.
In Chiang Mai, the tested Airalo Asialink regional configuration behaved more consistently than the global configuration during parts of the VPN-plus-video workload.
These results describe the specific sessions documented in the original Alertify research.
They do not establish a permanent ranking of eSIM providers or prove that the same relative performance would occur on another day, device or local network.
That distinction is especially important when testing mobile connectivity. Radio conditions, network congestion, routing, application behavior and device configuration can all influence the final result.
The real problem is not one application. It's competition between them.
During a conventional speed test, the connection is usually measured over a short interval using dedicated traffic to a selected test endpoint.
A working day looks different.
Real-time communication, file transfers, background applications and VPN traffic may all compete for limited available bandwidth and network resources.
This is particularly relevant when a phone is sharing its mobile connection with a laptop.
Need timely packet delivery, consistent upload capacity and low variation in delay.
Can introduce additional processing and another network path.
Can consume upload bandwidth without the user noticing how much traffic is generated.
Requires continuous upstream delivery and exposes instability over a longer period.
The original test setup used an iPhone 15 Pro and MacBook Air M2, with the laptop connected through the iPhone's Personal Hotspot.
The workloads included Zoom, Slack, Notion, Google Drive, Dropbox, VPN connections and a demanding live-streaming scenario using OBS Studio and YouTube Live.
Apple's documentation confirms that Personal Hotspot shares the phone's cellular data connection with another device. That makes the setup realistic, but it also introduces another element that future tests should control: the local hotspot connection itself.
A rigorous study should distinguish cellular-network limitations from local Wi-Fi hotspot effects, application behavior and device processing load.
Four locations. Four different ways performance deteriorated.
Explore the findings from the original Alertify Stress Test. Each location illustrates a different pressure point in mobile connectivity.
Medellín: when a good ping stops being a useful predictor.
What the original study recorded
Several global eSIM connections initially recorded latency below 70 ms. That suggested acceptable responsiveness during the lighter observation period.
Once video conferencing and VPN traffic operated together, some sessions exceeded 180 ms latency, with jitter rising above 30 ms.
Airhub's Latin America regional plan, connected through Movistar during the observed sessions, showed fewer severe spikes. Some Ubigi peak-period sessions were less consistent.
These observations reveal deterioration under load. They do not identify the specific network mechanism responsible.
Chiang Mai: the VPN exposed differences between products.
What the original study recorded
Enabling NordVPN added approximately 20–30 ms of latency during the observed sessions.
Under the combined VPN and video workload, some global Nomad sessions recorded packet loss exceeding 2%. Airalo's global configuration also showed instability in parts of the test.
Airalo's Asialink regional configuration behaved more consistently during the compared observations, with fewer visible interruptions.
The result is plan-specific. It does not establish that regional eSIMs universally outperform global plans.
Istanbul: the invisible cost of background uploads.
What the original study recorded
A roughly 500 MB Google Drive folder was synchronized while a Zoom screen-sharing session remained active.
Parts of the Nomad and Airalo global plan observations showed brief freezes and dropped video frames.
During the observed Ubigi session connected through Turkcell, latency generally remained around 40–60 ms without obvious jitter-related interruptions.
The study illustrates how an upload can disturb an otherwise usable connection. It does not prove that any particular provider has permanently superior upload management.
Athens: when the workload becomes a live production.
What the original study recorded
The test combined a 30-minute YouTube Live broadcast with Slack, VPN traffic and Google Drive activity.
Nomad's European regional plan disconnected from the stream twice.
Airhub's regional European plan completed the session with some dropped frames but without a full stream disconnect.
The observed Ubigi session completed with no dropped frames reported by OBS.
These outcomes describe the documented sessions. They are not repeated-trial success rates and should not be turned into a universal provider ranking.
The explorer presents observations from the original Alertify Stress Test. It does not generate simulated network results or establish current provider performance.
What the observations actually establish.
The original article combines numerical observations with qualitative assessments of application behavior.
The table preserves that distinction. Measurements are reproduced only where the original report provides them.
| Location | Concurrent workload | Published observation | What it demonstrates |
|---|---|---|---|
| Medellín Colombia | Zoom and VPN traffic. |
NUMERICAL OBSERVATION
Some connections moved from below 70 ms idle latency to above 180 ms under load. Jitter exceeded 30 ms during weaker sessions. |
Responsive performance without load does not guarantee responsiveness when applications compete. |
| Chiang Mai Thailand | VPN and video conferencing. |
NUMERICAL OBSERVATION
NordVPN added approximately 20–30 ms in observed sessions. Some Nomad global-plan sessions recorded packet loss above 2%. |
Additional traffic paths and simultaneous applications can reveal differences between individual products. |
| Istanbul Turkey | Zoom screen sharing with roughly 500 MB cloud synchronization. |
MIXED OBSERVATION
Brief freezes appeared in some global-plan sessions. The observed Ubigi session generally remained around 40–60 ms latency. |
Background transfers can coincide with deterioration in real-time applications. |
| Athens Greece | 30-minute live stream with VPN, Slack and cloud activity. |
SESSION OUTCOME
Nomad: two stream disconnects. Airhub: dropped frames, no full disconnect. Ubigi: no OBS dropped frames reported. |
Sustained upstream demand can expose instability not visible during short speed tests. |
Source: Alertify Research #03, The eSIM Stress Test, original publication. These observations are not normalized across identical radio, routing, plan, time-of-day and signal conditions. They support case-specific conclusions, not a statistical provider ranking.
A connection can have enough bandwidth and still become unresponsive.
The most important technical distinction in this research is between available capacity and responsiveness under load.
A connection may have sufficient download speed for individual applications, yet deliver inconsistent results when many data transfers happen at once.
This can occur when packets compete for a constrained part of the network. Traffic may accumulate in queues, increasing the time packets spend waiting before transmission.
The result can be higher latency, variation in delay and, when congestion becomes severe, packet loss.
The IETF discusses this behavior in RFC 9743, describing how excessive network queues—commonly called bufferbloat—can negatively affect interactive applications.
That technical principle helps explain why a connection can feel responsive during light browsing and much less responsive when another application begins transferring large amounts of data.
It does not prove that bufferbloat caused any specific deterioration in the original Alertify sessions.
Establishing the actual cause would require more detailed simultaneous measurements of throughput, queueing delay, network path, radio conditions and application traffic.
Why uploads can be particularly disruptive.
Consider a laptop connected through a phone's Personal Hotspot.
The traveler is presenting on Zoom. At the same time, a cloud application starts uploading a large folder.
Both applications need upstream capacity. If the combined demand approaches or exceeds the available capacity, packets may have to wait longer before transmission.
This can affect the timing of video and audio packets even though the connection itself remains active.
The user's immediate impression is that Zoom is unstable. The wider network problem may involve simultaneous traffic that Zoom does not control.
Video and audio require consistent upstream and downstream delivery.
Another application begins competing for available upload capacity.
Congestion and queueing can make packet delivery less predictable.
Audio, video or screen-sharing quality may deteriorate.
Why the upload result matters more than it first appears.
Travel eSIM comparisons commonly emphasize download speed.
That's understandable. Download speed is easy to explain and relevant to browsing, streaming and downloading.
But remote work creates a different pattern of demand.
Video calls, screen sharing, cloud backups, file transfers and content publishing all require upstream capacity.
A connection that looks fast during a download test may behave very differently when a large upload competes with interactive communication.
This is why the Istanbul and Athens observations deserve particular attention.
They show what happens when the traveler stops consuming content and starts producing it.
A VPN changes the path. It doesn't explain every slowdown.
Virtual private networks are widely used by business travelers and remote professionals.
They can protect traffic between the device and a VPN endpoint and support access to corporate services.
They can also change the route taken by application traffic and introduce additional processing.
In the original Chiang Mai observations, enabling NordVPN added approximately 20–30 ms of latency during the measured sessions.
That is an observation about those particular sessions, not a universal performance cost of VPN usage.
The effect of a VPN depends on the selected server, tunneling protocol, physical distance, device performance and network conditions.
It is also important to distinguish the VPN's own effect from problems caused by concurrent traffic.
If a call becomes unstable while a VPN and cloud synchronization are active, the cause cannot be assigned to the VPN without testing the components separately.
The most useful approach is to compare the same workload under controlled conditions, first without the VPN and then with a documented VPN endpoint and protocol.
Why regional and global eSIM differences deserve investigation.
The Chiang Mai observations also showed different behavior between Airalo's regional Asialink configuration and a global configuration.
That is a valuable finding because it challenges the assumption that every plan sold by one provider uses an identical connectivity setup.
Different products may involve different network arrangements, upstream suppliers or policies.
However, the original stress test does not publish enough network-path evidence to establish exactly which architectural difference produced the observed result.
This is the research question explored further in Research #02 — The eSIM Routing Report .
The metric that matters is whether the work can continue.
Network measurements are essential for understanding performance. But they should not become the only definition of success.
A traveler does not purchase an eSIM because they enjoy measuring milliseconds.
They purchase connectivity to accomplish something.
Join a meeting. Access a corporate application. Upload a presentation. Publish a video. Deliver work on time.
The quality of that experience depends on how the network supports the actual application.
Zoom already measures more than speed.
Zoom's own statistics distinguish latency, jitter, packet loss, resolution, frame rate and bandwidth.
Its documentation recommends, as general application guidance, latency of around 150 ms or less, jitter of 40 ms or less and packet loss of 2% or less.
These are Zoom's application-oriented recommendations, not universal telecommunications quality thresholds.
The original Medellín observations, where some loaded latency measurements exceeded 180 ms, illustrate why higher delay deserves attention during live communication.
But the study's network measurements should not be treated as though they were necessarily captured from Zoom's internal statistics.
They are different measurement contexts unless the original logs establish otherwise.
Video quality depends on competing traffic.
The IETF's WebRTC media transport standard explains that interactive media must account for the available capacity of the network path and competition from other traffic.
When the connection cannot sustain the required traffic, interactive applications may need to adapt.
That adaptation can preserve a usable conversation, but it can also reduce video quality or create interruptions.
A useful eSIM benchmark should therefore report both network conditions and application outcomes.
OBS results reveal a different kind of failure.
Live streaming requires sustained upstream delivery to a remote ingest server.
OBS Studio's official documentation explains that network-related dropped frames can occur when the connection cannot maintain the configured streaming bitrate.
OBS also distinguishes network dropped frames from skipped frames caused by encoding problems and frames missed because of rendering delays.
Those distinctions matter when interpreting a stress test.
The Athens findings are meaningful as original session observations. But a future benchmark should preserve the actual OBS logs and separate network drops from device-side encoding or rendering issues.
A network benchmark measures the connection. A workload benchmark measures whether the connection supports the activity the user needs to complete.
Slower performance under load is not automatically throttling.
The original research also described deterioration during some high-data-use sessions.
Symptoms included slower uploads, increased latency and unstable VPN behavior.
Those symptoms are important, but they are not sufficient to prove that a provider deliberately reduced speed.
Several different mechanisms can produce similar results.
- Network congestion: demand for shared network capacity increases, affecting performance.
- Radio conditions: signal quality or interference reduces effective throughput.
- Routing variation: the traffic path changes or introduces additional delay.
- Plan-level traffic management: a documented policy reduces available throughput after defined conditions are met.
- Application or device limitations: the VPN, hotspot, encoder or software introduces another bottleneck.
Distinguishing these mechanisms requires more than comparing one speed test before and after a large upload.
It requires repeated measurements, consistent network conditions and the applicable plan rules.
That is also why the Stress Test should remain distinct from The Unlimited eSIM Reality Test.
The Stress Test investigates competing applications and sustained workload behavior.
The Unlimited Reality Test investigates disclosed high-speed allowances, reset rules and post-threshold usability.
Both matter, but they ask different research questions.
Introducing a more useful definition of eSIM workload performance.
The next step for Alertify is to make loaded performance as visible as download speed.
That doesn't require inventing a new universal performance score.
It requires measuring the right things consistently.
Latency under load
The difference between an idle connection's responsiveness and responsiveness while applications are competing for capacity.
A future test should collect repeated latency measurements during idle periods, downloads, uploads and combined workloads.
It should report medians, upper-percentile values and the degree of variation, not just one average.
Sustained upload stability
A short upload-speed measurement is useful but may not reflect a prolonged video stream or cloud synchronization session.
Sustained tests should document successful transfer rate, interruptions, retransmissions where observable, and the ability to maintain interactive traffic at the same time.
Workload completion
A meaningful test needs clear success criteria for the applications being used.
For example, a video-call scenario might measure whether audio remained understandable, whether the application reported packet loss and whether screen sharing was interrupted.
A streaming scenario might measure whether the session remained connected and whether OBS reported network drops.
Baseline response time measured before the demanding applications begin transferring traffic.
Response time during defined concurrent workloads, including uploads and downloads.
Variability in delay while the network is serving multiple competing applications.
Lost test packets during sustained use, reported with measurement method, endpoint and sample size.
Ability to sustain upstream traffic over time without unacceptable interruptions.
Whether a defined video call, transfer, VPN operation or stream completes successfully.
How quickly responsiveness returns after the competing workload has stopped.
Whether similar outcomes appear across multiple sessions, times and comparable network conditions.
These measures would allow Alertify to distinguish three important connection states.
A connection can be technically available, responsive during ordinary activity and capable of sustaining demanding work.
Those are not necessarily the same achievement.
Not every traveler needs a stress-tested connection. Some absolutely do.
The point of this research is not that every customer should buy the strongest possible connectivity product.
Someone using maps, messaging and casual browsing may never encounter the demanding conditions examined here.
For that traveler, an affordable, dependable plan may offer excellent value.
The calculation changes when connectivity becomes part of the working environment.
The meeting matters more than the speed-test screenshot.
Executives, consultants and sales professionals need stable conferencing, VPN access and collaboration while moving between destinations.
A normal working day creates abnormal mobile demand.
Cloud synchronization, remote development, multiple devices and prolonged video sessions can expose weaknesses that light travel use would never reveal.
Upload performance can determine whether work gets delivered.
Publishing video, sending large media files and live production depend on stable upstream traffic rather than download performance alone.
What practical users can do.
Before an important call, a traveler should test the actual working setup: the same eSIM, hotspot, VPN and applications they expect to use.
Where practical, large cloud transfers can be paused during critical meetings.
A traveler can also reduce unnecessary background activity and check whether their provider permits hotspot sharing on the selected plan.
These precautions cannot eliminate radio or network congestion, but they can reduce avoidable competition for a constrained connection.
For work that cannot tolerate extended disruption, a separate backup connection may also be appropriate.
Travel eSIM suppliers should be evaluated by workload, not just coverage.
The research has implications beyond consumer eSIM purchasing.
Airlines, travel platforms, banks, hotels and loyalty programs can integrate connectivity into their customer relationships.
Enterprise mobility providers must also consider how international connectivity behaves when employees use business applications.
Many supplier comparisons emphasize country coverage, wholesale price, available networks, activation and API capabilities.
Those are necessary considerations. But they do not establish whether a product remains usable during sustained professional work.
A coverage agreement is not a workload guarantee.
A provider may offer access to a large number of mobile networks.
That does not automatically establish performance for video conferencing, VPN use or large cloud transfers.
The specific plan, visited operator, traffic path and network conditions remain important.
B2B buyers should therefore investigate how the intended connectivity product performs in priority destinations.
The economics of failure are not limited to lost data.
When a travel eSIM becomes unstable, the customer may not run out of mobile data.
Instead, a meeting may be interrupted, an upload may fail, or the traveler may need to switch to another service.
These are potential operational costs, although the original Stress Test did not quantify their financial impact.
That distinction matters for organizations purchasing connectivity for workforces or distributing it to demanding customer segments.
The better question is not simply whether the service provides mobile internet.
It is whether that internet connection supports the activities for which the customer purchased it.
Methodology and limitations
This edition reconstructs and expands the original Alertify eSIM Stress Test, Research #03.
The research observations come from the previously published article. The technical explanations and testing recommendations are editorial additions.
No new live network measurements have been conducted or claimed for this editorial revision.
Chiang Mai, Thailand; Medellín, Colombia; Istanbul, Turkey; and Athens, Greece.
iPhone 15 Pro and MacBook Air M2, with the MacBook connected through the iPhone's Personal Hotspot.
Zoom with video and screen sharing, Slack, Notion, Google Drive, Dropbox, VPN connections and YouTube Live streaming.
The original article identifies Speedcheck Pro, PingPlotter and OBS Studio logs as observation tools.
Selected latency and jitter observations, packet-loss figures, application interruptions, dropped frames and streaming outcomes.
Original field observations, numerical examples and qualitative findings are distinguished from technical interpretation and future methodology.
The original publication does not provide complete raw session logs, exact purchased plan identifiers for every observation, controlled signal measurements, all VPN endpoint details, comprehensive OBS log exports or repeated-trial counts. Consequently, the findings cannot be used to calculate statistically robust provider rankings, prove intentional throttling, identify network priority classes or isolate the technical cause of every interruption.
How the next edition should be tested
A future controlled Stress Test should compare precisely identified plans using repeatable traffic patterns and documented network conditions.
Testing should begin with a baseline, then introduce demanding traffic in clearly defined stages: video conferencing, VPN use, cloud synchronization and sustained streaming.
The protocol should measure idle and loaded latency simultaneously with the actual applications.
It should also record throughput, packet loss, jitter, application interruptions and recovery after the competing traffic stops.
A robust design would repeat measurements across multiple time periods and days, documenting signal conditions, device software, selected operator and measurement endpoints.
Where hotspot use is involved, the test should control the local device-to-phone connection and distinguish its effects from the cellular network.
For OBS tests, network dropped frames should be distinguished from rendering and encoding problems.
Tests must also respect provider terms, including traffic-management and acceptable-use conditions.
Publishing the measurement methodology, sample sizes, raw or suitably summarized results and unresolved observations would make future editions more reproducible and easier to evaluate.
Original research and technical sources
The provider and destination observations come from Alertify's original Stress Test. The additional references explain the technical mechanisms relevant to concurrent network traffic.
The real stress test begins when the speed test ends.
The eSIM Stress Test investigates a dimension of mobile connectivity that conventional comparisons frequently overlook: performance under simultaneous demand.
The original observations from Medellín, Chiang Mai, Istanbul and Athens demonstrate how a usable mobile connection can deteriorate when video calls, VPN traffic, cloud synchronization and sustained uploads compete for available capacity.
The findings also show why a single provider ranking would be too simplistic.
The tested product, destination, visited network, time and workload can all influence the outcome.
A regional plan may perform more consistently than a global configuration in one location.
Another product may remain stable during an upload-heavy session but behave differently elsewhere.
These observations are valuable without being converted into claims of universal superiority.
The broader lesson is that connectivity quality is dynamic.
A connection suitable for maps and occasional browsing may not be suitable for prolonged professional workloads.
A connection with strong advertised download speeds may still experience high latency under load or unstable upstream delivery.
And a connection with less impressive headline numbers may prove more useful when it remains responsive during demanding work.
The next generation of travel eSIM benchmarking should measure not only whether data can move, but whether important work can continue while it does.
That is the difference between measuring connectivity and measuring its usefulness.