GeoFINDR: Practical Approach to Verify Cloud Instances Geolocation in Multicloud
Saïd Ider, Maryline Laurent · 2025
Read GeoFINDR: Practical Approach to Verify Cloud Instances Geolocation in Multicloud by Saïd Ider, Maryline Laurent (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-18685.txt.
Introduction
Saïd Ider and Maryline Laurent check whether a cloud server is really where its provider claims, by pinging it from the outside world, in this 2025 paper. Location matters legally and technically, and a provider might lie. Their GeoFINDR system locates a virtual machine using the RIPE Atlas network of measurement probes, under a threat model that assumes the lie.
Four ideas make it practical. The measurement starts from inside the suspect machine itself. A greedy algorithm picks a first ring of landmarks near the declared location. A sectorization step then finds other landmarks whose delay behavior matches the machine's, narrowing the compass sector it hides in. The final estimate is the weighted center of that matching set. Experiments reached accuracy as fine as 22.1 kilometers even against a lying provider. The tool is published as open source for auditors and customers alike.
Catalog record
| CREATOR | Saïd Ider, Maryline Laurent |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-18685
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
GeoFINDR: Practical Approach to Verify Cloud Instances Geolocation in Multicloud Thanks: This work was supported by the France 2030 ANR Project ANR-23-PECL-0009 TRUSTINCloudS. Saïd Ider, Maryline Laurent Abstract In multicloud environments, where legal obligations, technical constraints and economic interests are at stake, it is of interest to stakeholders to be able to locate cloud data or the cloud instance where data are decrypted for processing, making it particularly vulnerable. This paper proposes an original and practical delay-based approach, called GeoFINDR, to locate a cloud instance, e.g. a Virtual Machine (VM), over the Internet, based on RIPE Atlas landmarks. First, the assumed threat model and assumptions are more realistic than in existing solutions, e.g. VM-scale localization in multicloud environments, a Cloud Service Provider (CSP) lying about the VM’s location. Second, the originality of the approach lies in four original ideas: (1) geolocalization is performed from the VM, (2) a Greedy algorithm selects a first set L M A LM_{A} of distributed audit landmarks in the vicinity of the declared area, (3) a sectorization algorithm identifies a set L M S LM_{S} of other landmarks with distance delay behavior similar to that of the VM to estimate the sector of the VM, and (4) the estimated location of the VM is calculated as the barycenter position of the L M S LM_{S} landmarks. An open source tool is published on GitHub and experiments show that localization accuracy can be as high as 22.1 k m 22.1km , under unfavorable conditions where the CSP lies about the location of the VM. Keywords: Delay-based geolocation, multicloud, VM-scale localization, GeoFINDR, RIPE Atlas, dishonest cloud service providers 1 Introduction Even if customers specify in their SLAs that they want their data and data processing to be located in data centers in certain regions or cities, it is difficult for them to be sure that their SLA is being met, because most cloud services operate like a complex black box. Verifying the actual location of data and workloads from the declared cloud region is virtually impossible and never 100% reliable anyway, even when expensive means are deployed. Geolocalization of data and physical resources on the Internet is very useful [ 1 ] [ 2 ] for customers and also for cloud service providers (CSP) themselves, in meeting their obligations to customers in multicloud environments and avoiding penalties. (1) For legal and sovereignty reasons, there is a need to have guarantees where the externalized data is stored to benefit from GDPR protection [ 5 ] , to avoid the CLOUD Act, which allows the US to collect data from US companies or data stored in US cloud servers. (2) For availability and resiliency reasons, customers have an interest in ensuring that their virtual resources are replicated in different geographic locations to avoid, for example, losing all their data in the event of a fire, as happened in 2021, resulting in OVH paying customers at least 400,000€ in compensation 1 1 1 https://www.transatlantic-lawyer.com/ovh-must-pay-more-than-400000-e-after-a-fire-destroyed-its-data-centers-why-this-decision-is-important-for-hosting-providers-hosting-eu-personal-data/ . (3) To combat mistrust of CSP, it is important to provide guarantees on the location of virtual resources. This paper proposes a new approach called GeoFINDR, to verify the physical location of cloud instances, e.g. a VM, over the Internet. GeoFINDR is practical in many aspects. It supports VM-scale localization, which has a higher granularity than the cloud server loacalization, which is provided by most of the geolocation solutions of the literature. Localizing at VM scale is interesting as it matches the requirements of multicloud environments and the VM is known as a vulnerable instance, since it is there that data is decrypted before being processed. An open source tool is available on GitHub 2 2 2 https://github.com/diasredi/GeoFINDR , and experiments show upto 22.1km accuracy with an average of 22.6km , which are very good results compared to other solutions. Note that these measures were obtained under unfavorable conditions where the CSP lies about the location of the VM. The paper is organized as follows. After positioning our GeoFINDR solution against related works on delay-based solutions in Section 2 , Section 3 presents our GeoFINDR approach, from the threat model and assumptions to the improvements brought to the area of delay-based solutions. Section 4 describes the implementation details, including the resulting pseudocode. Section 5 presents the experiments performed with the methodology and fully explains the achieved results, before concluding in Section 6 . 2 Related Works on Delay-based Solutions Delay-based solutions rely on multilateration to locate an entity, without CSP intervention, and derive the position of an entity from a set of landmarks with known positions and a method for calculating the distance between the entity and the landmarks. Table 1 summarizes the main features of the existing solutions. CBDG (Constraint-Based Data Geolocation) [ 9 ] is the first software implementation for locating data over the Internet. CBDG uses PlanetLab servers as landmarks and calculates the Distance-Delay Relation (DDR) by linear regression of the ping RTT (Round Trip Time) between all landmarks. All landmarks then ping the server to measure RTT and derive a distance that is used to define areas where the server might be located. We emphasize that attempting to locate data is pointless, since data is duplicated anyway due to the high-availability architecture of the cloud and can circulate by physical means. Therefore, we consider the issue of data location outside the scope of our study and instead focus our efforts on verifying the location of a cloud instance, which is relevant because it is where encrypted data is processed. In IGOD [ 10 ] , the authors improve CBDG DDR computation, overall performance, and accuracy by implementing several refinements, such as using file transfers instead of pings, and computing a minimum regression instead of an average regression. In addition, IGOD introduces a while loop that selects three dispersed landmarks in an area where the server is located, which gets smaller until it reaches a small enough satisfactory size. They showed that dispersed and closer landmarks gave more accurate results than more distant and randomly chosen landmarks. In doing so, they significantly reduced the number of landmarks needed to estimate the server’s position compared to CBDG with greater accuracy. In SPLITTER [ 12 ] , the authors use a ”double ring triangulation” to reduce the area in which the server could be located, and they manage to reduce the uncertainty of an estimate, but they do not address common issues like the risk of zero intersections. In VLOC [ 13 ] , the approach to server localization is reversed from previous solutions. Instead of performing the audit from outside the server, they propose a method for each VM to perform the audit internally using available landmarks. VLOC use web pages as landmarks with approximate positions due to low reliability of IP databases [ 14 ] . DDR is defined by average delays of measurements with landmarks combined with a trust coefficient to be specified. The advantage of internal audit, as proposed by VLOC and our GeoFINDR approach, is the ability to localize at the cloud instance level, while the external audit limits location to a single server instance or cloud PROXY. Therefore, internal audit allows for better