How Cloud Edge Infrastructure Improves Round-Trip Time

Today, cloud providers offer various services to customers, including storage, computation, and networking. To facilitate this, they deploy cloud data centres (cloud regions), typically located in large metropolitan areas, which serve as the centralised backbone of their services (e.g., AWS regions). Since these cloud data centres are often distant from end-users, cloud providers also establish cloud edge locations (e.g., AWS Local Zones) to reduce distance and latency. These edge locations are geographically dispersed, enabling reduced latency, but are smaller, and provide a limited range of services.

The work of Martin and Dogar¹ spanned per continent showed that cloud edge infrastructure could improve Round Trip Time (RTT) by up to 55% compared to cloud regions offered by Amazon, Google, and Azure. In Europe, RTTs decreased on average from 17.64ms to 12.9ms, a 26% improvement using the cloud edge instead of cloud regions. However, does this improvement apply to all European countries?

We build upon their findings, narrowing the focus to Europe at a country level. Martin and Dogar deployed Virtual Machines (VMs) in each cloud region of Amazon, Google, and Azure, to measure the RTT from RIPE Atlas probes, without the use of the cloud edge. We categorise the countries into two groups according to the availability of cloud regions (VMs), geographic diversity, and strategic significance. Germany, Finland, and Poland have VMs located in local cloud regions, while Bulgaria, Greece, and Czechia do not. Furthermore, these countries are a sample of European countries from diverse parts, covering the north, south, and central regions.

Results from our work show that the use of cloud edge infrastructure in Germany saw a 17% speedup, in Finland 24%, Poland 18%, Greece 68%, Bulgaria 75%, and Czechia 50%, compared to cloud regions. These results reveal significant disparities in RTT improvements across Europe; countries like Greece, Bulgaria, and Czechia benefit far more from cloud edge infrastructure, achieving improvements well above the 26% on average.

Dataset and methodology

Martin and Dogar conducted a study leveraging RIPE Atlas built-in measurements (pings)² to analyse RTTs across continents from 9–13 May, 2023. Their research focused on comparing RTTs to cloud regions and edge locations of three major cloud providers (AWS, Google, and Azure). The data was sourced from globally distributed probes measuring latency to cloud regions and cloud edge infrastructures. To facilitate comparison, they introduced two key metrics: the baseline and the edge.

For the baseline, 100 virtual machines (VMs) were launched in each cloud region of the three major cloud providers (AWS, Google, Azure), with globally distributed probes pinging these VMs to record the minimum RTTs.

For the edge (cloud edge infrastructure) measurements, anycast IP addresses provided by routing optimisation services on each cloud provider were used to determine the minimum time to reach their private WAN.

In our work, we focus exclusively on VMs and probes located in the six selected European countries. Furthermore, to better understand the impact of distance from a VM in RTTs we geolocate the sources (probes distributed globally) and targets (VMs) of the measurements using the MaxMind database.³ It is unnecessary to geolocate the target cloud edge IPs for edge measurements, as cloud edge infrastructures are distributed across Europe. Since these IPs are anycasted, the target cloud edge infrastructure is automatically the one closest to the probe initiating the edge measurement.

Results

In this section, we acknowledge that our concerns regarding the slight improvements to the baseline with cloud edge solutions in Europe do not apply uniformly across all countries. Notably, countries such as Greece, Bulgaria, and Czechia are experiencing significant enhancements in their RTTs thanks to the implementation of cloud edge technology.

RTT improvements with cloud edge at the country level

In our analysis, we notice that the distance of the probes of the country from VMs affects the RTTs for the baseline measurements, so we categorise the six European countries into those with and without VMs to analyse them separately. Figure 1 depicts the VMs that are deployed in Europe. From the European countries offering VMs, we opted for Germany, Finland, and Poland, positioned in central and northern Europe. In regions lacking VM availability, we selected Greece, Bulgaria, and Czechia. Greece and Bulgaria hold strategic geopolitical importance at the crossroads of Europe, Africa, and Asia, offering valuable insights for a wide range of organisations. Additionally, Czechia, situated in central Europe, shares proximity with countries hosting VMs. This adjacency facilitates connectivity and underscores the importance of distance from a VM in the baseline analysis.

Figure 2 depicts the comparison of baseline and edge RTTs for the selected countries with and without VMs. Particularly, the countries with VMs shown in Figure 2a consistently achieve RTTs of under 20 ms for 80% of probes in both edge and baseline scenarios. However, the good RTTs in baseline for Germany and Poland are not solely due to the existence of VMs in those countries, but also to the fact that they are close to many other VMs. A special case is Finland, which achieves low RTTs despite not having VMs close to it, except its own. This indicates that Finland has very good Internet connectivity to all of Europe.

On the other hand, in regions where VMs were not deployed (Greece, Czechia, Bulgaria), the edge solution demonstrates a significant performance advantage over the baseline Figure 2b. In comparison to the baseline, the edge achieves a lower latency of approximately 50-70%. What is even more interesting is that the closer a country is to another one that has a VM, the performance difference between baseline measurements is increasing. As Figure 1 shows Czechia is closer to VM than Bulgaria and Bulgaria is closer than Greece and thus achieves lower RTTs. Table 1 summarises the above results.

Case study: Italy

To further investigate if the existence of VM in a country has an impact on baseline measurements, we focused on Italy. The reason behind the selection is that although there are probes distributed all over the country, as Figure 3a depicts, VMs are only deployed in the northern part, as depicted in Figure 3b. We compare the baseline with the edge RTTs of northern, central, and southern parts of Italy.

Given Figure 4 we can observe that RTTs for the baseline measurements for the two Italian south probes are the worst because all VMs are far away from them. Instead, the Italian north and central baseline measurements are better due to being closer to the rest of the VMs, suggesting that the above assumption is valid. As for the Italian probes, the edge follows the same distribution as the previous countries, except for the southern probes where the edge is better than the baseline, but the RTTs are still large. The reason behind this is not clear, we need more indicators to clarify it.

Limitations and conclusions

Our work has several limitations. The measurements were conducted over a period spanning four days in May 2023, which may not capture broader temporal trends. Geolocating vantage points relied on MaxMind, but the lack of historical data required us to use geolocation information from 2024, potentially introducing inaccuracies. Additionally, a limited number of VMs constrained our ability to perform comprehensive baseline measurements. These factors may affect the precision of our results.

Our approach focuses on the comparison of latency in six European countries. Our results suggest that for all those countries the edge has better performance in terms of latency than baseline. Also, the distance of the countries from VMs plays a crucial role in our latency analysis. For the baseline, the closer a country is to a VM, the more similar the latency of edge and baseline becomes. Moreover, the countries that are far away from VMs benefited more from the cloud edge, the RTTs for Greece, Bulgaria, and Czechia improved 68%, 75%, and 51% respectively for the 80% of probes. To further validate our assumptions we investigated a special case in Italy.

Future work

As mentioned before, our work has a lot of limitations. So part of future work is to include more European countries and add more VMs, making sure that we have a complete view of the impact of cloud edge for each country. Finally, we could even investigate more and understand on a city scale what is the impact, and results that might be more useful for the Internet/research community. This information could potentially be useful for various fields/applications (e.g. Smart cities).

References

1. Noah Martin and Fahad Dogar. 2023. Divided at the Edge - Measuring Performance and the Digital Divide of Cloud Edge Data Centers. Proc. ACM Netw. 1, CoNEXT3, Article 16 (December 2023), 23 pages. https://doi.org/10.1145/3629138
2. https://atlas.ripe.net/docs/built-in-measurements/#accessing-the-results
3. https://stat.ripe.net/docs/02.data-api/maxmind-geo-lite.html