The Prisma Cloud Connectivity Report 2024 analyzes network latency between AWS regions and Cloudflare PoPs, identifying regions with the fastest and slowest connectivity. It highlights the impact of physical infrastructure and geographical distance on latency, providing insights for optimizing cloud connectivity and data exchange efficiency.
At Prisma, we are operating services in most AWS regions and all Cloudflare PoPs outside of China. As a result, we have an extensive set of latency data for requests between Cloudflare and AWS. This question on X prompted us to start releasing this data in a yearly Cloud Connectivity Report:
We collect p50, p90, p95, p99 and p999 latency metrics for the intersection of 16 AWS regions and 266 Cloudflare points of presence (PoPs) for a total of 21k data points. This report will focus on a few key data points, and the full dataset is linked at the end.
AWS regions with the fastest connectivity to Cloudflare
The three AWS regions with the lowest latency to a nearby Cloudflare PoP are all in Asia Pacific:
Fastest of all, ap-south-1 (Mumbai) opened in 2016 and has a p50 latency to the nearby Cloudflare BOM PoP of 10.9ms. Second, ap-southeast-1 (Singapore) opened in 2010 and has a p50-latency to the nearby SIN PoP of 12.3 ms. Finally, the ap-southeast-2 (Sydney) region opened in 2012 and has a p50 latency of 15.8ms to the nearby SYD PoP.
To understand why some AWS regions have really low latency access to a Cloudflare PoP, all you need to do is follow the wire. At the end of the day, these are physical networks with fiber optics connecting them at an Interconnection Facility. As an example, the ap-south-1 (Mumbai) AWS region is peering at three facilities:
The Cloudflare BOM (Mumbai) PoP is also peering the Equinix MB1 facility, so the network distance between the Cloudflare and AWS facilities in Mumbai is very short:
Contrast this with the us-east-2 AWS region in Columbus, Ohio which has a p50 that is almost three times as high to the nearest Cloudflare PoP CMH, which is also in Columbus:
us-east-2 is peering at just a single interconnection facility:
Cloudflare also has a PoP in Columbus, but it is peering at a different Cologix facility, resulting in additional network hops:
Latency maps for popular AWS regions
A benefit of the Cloudflare edge network is that they have PoPs in many cities. The largest AWS regions have low latency connections to several Cloudflare PoPs.
us-east-1
eu-central-1
High latency connections
Prisma Accelerate is a global database cache, increasing the performance of websites that are deployed in multiple regions and communicating with a central database. As such, we see traffic from almost all Cloudflare PoPs to any given AWS region. These are the AWS region and Cloudflare PoP pairs where we observe the highest latency:
It’s no surprise that these are connections over a very long distance. Buenos Aires to Ireland, Hong Kong to Sweden, and Johannesburg to Sydney.
On the Prisma Accelerate network we are able to serve many of these requests from cache, directly from the Cloudflare PoP, making sure our customers’ applications feel snappy to their end users. Not all requests can be cached, so we are working with our providers to ensure all requests happen over the fastest route.
We’ll report on the latest state in the Prisma Cloud Connectivity Report next year. Let us know via X or Discord if there are other data points you’d like to see!
You can find the full dataset as a CSV here: aws-cf-latency-2024.csv
Don’t miss the next post!
Sign up for the Prisma Newsletter
