LACNIC Blog
> Research > Measuring Response Times to the LACTLD Anycast Cloud
Measuring Response Times to the LACTLD Anycast Cloud
26/01/2022
By Hugo Salgado, Research & Development Engineer at NIC Chile
Popular belief holds that one cannot understand what one cannot measure, and the LACTLD Anycast Cloud Response Time Study conducted by LACNIC’s R&D department proves the saying to be true.
The study provides monthly measurements. Even after the short time it has been running, it has already produced relevant information regarding response times from Latin America which is helping to find where improvements can be made and to plan the next steps ahead.
About the LACTLD Anycast Cloud
The LACTLD Anycast Cloud is a free DNS service for DNS zones in LACTLD member countries. It includes an authoritative server which contains the DNS zones of more than 14 countries in the LAC region and helps to strengthen each country’s existing infrastructure, providing a service operated by the LACTLD community itself. In addition, this service uses a technology known as “anycast” which allows maintaining copies in different locations, thus reducing cloud response times for each country.
The DNS zones of certain countries are hosted in the anycast cloud. In turn, each node can be hosted by collaborating ISPs and Internet exchange points who volunteer to do so. All of this brings the information closer to each network. Currently, the LATLD Anycast Cloud has more than 15 nodes with exact copies of the service hosted in more than 10 countries. Each of these copies provides additional resilience to the service, as they preferentially respond to the closest networks and countries. They can also be reorganized and become a backup for other distant nodes in the event of an outage or saturation. This means that each node operates and is activated as a backup for another, resulting in enormous levels of robustness and resilience for LACTLD member countries. Likewise, hosted nodes improve robustness and resilience for end-clients in these countries.
For more information, go to https://anycast.lactld.org/
(Free access, no subscription required)
About LACNIC’S nsstats Project
LACNIC’s nsstats Project is a system developed and hosted by LACNIC which periodically measures the response times from each of the RIPE Atlas probes (https://atlas.ripe.net/) hosted in Latin America to the LACTLD Anycast Cloud. These probes are hosted by volunteers on residential networks and by some ISPs, and they perform on-demand Internet infrastructure measurements launched by researchers worldwide. LACNIC performs DNS measurements that send queries from these probes to the LACTLD Anycast Cloud, responses are provided by the nodes closest to each probe, and response times are recorded. This number tells us how fast the cloud service is from the end user’s point of view.
LACNIC’s nsstats Project is a system developed and hosted by LACNIC which periodically measures the response times from each of the RIPE Atlas probes (https://atlas.ripe.net/) hosted in Latin America to the LACTLD Anycast Cloud. These probes are hosted by volunteers on residential networks and by some ISPs, and they perform on-demand Internet infrastructure measurements launched by researchers worldwide. LACNIC performs DNS measurements that send queries from these probes to the LACTLD Anycast Cloud, responses are provided by the nodes closest to each probe, and response times are recorded. This number tells us how fast the cloud service is from the end user’s point of view.
The data obtained from the study can be used to analyze multiple perspectives on the use of the LACTLD Anycast Cloud.
On the one hand, all countries considered, the average response time is less than 185 ms. While this number may be too high for certain transactions, it is within the maximum permissible DNS query wait time, i.e., on average, one might consider that no query goes unanswered (there are no timeouts). On the other hand, response times from six countries are less than 50 ms, which is within the optimal numbers for the DNS. It is also no coincidence that the countries with the longest response times (slower responses) are precisely those that do not have any copies of the cloud within their networks. This shows the importance and effect of having nodes in each country.
A more detailed analysis also shows some cases that deviate from the norm. Despite having internal copies, some countries do not have good response times. This is generally due to internal routing issues within each country where, as a result of the internal policies of each Internet service provider, local copies are not being fully leveraged and, in certain locations, queries are being sent to copies in a different country instead of being processed locally. A more in-depth analysis of this is being conducted and the topic is being discussed with the institutions hosting these copies so that, in coordination with other ISPs and IXPs, they can improve route announcements and optimize their reach.
Additionally, a historical analysis of the past few months shows quite a bit of stability, generally attributable to minor improvements in some of the countries. Examples include Chile, which went from the fifth to the second shortest response time, mainly due to the fact that a second node was installed in the country in November. This node is hosted by PIT Chile, one of the country’s main Internet exchange points, and has benefited the country and nearby networks. This shows the impact of the efforts to participate in the LACTLD Anycast project.
Finally, it should be noted that the measurements and results that can be obtained depend on the tool that is used. Therefore, we invite institutions to collaborate with the LACTLD Anycast project by hosting anycast cloud copies, as well as RIPE Atlas probes to allow us to increase the number of measurement points in each city.
